Abstract The cryptocurrency market represents a significant innovation in the financial ecosystem, built upon cryptographic principles to ensure secure and transparent transactions. Cryptocurrencies experienced a global adoption, driven by their decentralized nature that enables borderless transactions without third-party intermediaries. The price of cryptocurrencies is characterized by a significant volatility, that introduces both opportunities and challenges. In this context, the development of accurate methods for the forecasting of price variation, able to work in real-time on data streams, has become vital for various stakeholders. In this paper, we propose a novel approach, called LEMON, for the online prediction of the price variation of cryptocurrencies, that leverages possible temporal correlations among them. Our approach stems from the empirical evidence that cryptocurrencies tend to form groups characterized by similar trends, a behavior often attributed to shared market dynamics and common external factors. Through the analysis of temporal correlations, LEMON dynamically identifies these groups, that are then exploited to learn multiple multi-target tree-based models, specifically designed for processing continuous data streams. LEMON also introduces a novel adaptive non-parametric weighting scheme, that automatically adjusts the importance of each instance based on the observed data distribution in real-time, improving the forecasting of the price variation. Our experiments, performed on 16 datasets related to 16 cryptocurrencies, demonstrate that LEMON outperforms state-of-the-art approaches in two distinct prediction tasks: forecasting the closing price variation (regression) and predicting the market trend direction (classification), making it an effective tool to support stakeholders requiring accurate real-time predictions.

Abstract The quality assessment of fruits and vegetables is crucial in the agroalimentary supply chain, as it directly affects consumer satisfaction, market value and overall food security. Traditional approaches rely on visual inspections or destructive techniques, which are labor-intensive and time-consuming. On the contrary, non-destructive techniques emerged as promising alternatives, offering solutions that can be adopted in real environments. Previous studies emphasized that the color distribution over images plays a significant role in the quality evaluation of food. In this paper, we propose a solution that leverages an autoencoder architecture to extract groups of relevant colors from the complete histogram of colors. To enhance the analysis of real-world images with complex backgrounds, we employ a pre-trained U2-Net architecture for background removal. Moreover, we propose a novel procedure based on outlier detection to identify and remove parts of the background that are not fully eliminated, especially along the edges of the product. After this preprocessing, we extract a complete color histogram which is fed to an autoencoder architecture, to extract high-level features representing color groups at different levels of granularity. The goal is to make the learned models less sensitive to light and color perturbations. Our experiments, conducted on two real-world datasets related to two different learning tasks, demonstrated the effectiveness of the proposed solution, that outperformed several baseline and state-of-the-art approaches, also based on complex neural network architectures.

The increasing sophistication in AI models makes Explainable AI (XAI) extremely relevant for enhancing predictions with explanation capabilities. Recent XAI approaches for tabular, image, and graph data have become popular and widely adopted. However, the multi-dimensional nature of spatio-temporal data collected in sensor networks makes most approaches ineffective in such scenarios. XAI approaches that specifically deal with this type of data are emerging, but the complexity of the extracted explanations leads to multi-axis information for nodes, features, and timesteps that is cumbersome to visualize. Indeed, providing effective visualizations could help bridging the existing gap between XAI model predictions and their fruitful and responsible exploitation in practical domains. In this paper, we address this gap by studying the effectiveness of different visualization techniques with multi-dimensional explanations. We adopt a meta-learning XAI framework that identifies salient factors from multiple analytical views. Then, we present a qualitative and quantitative study comparing the effectiveness of 14 visualization techniques using two real-world sensor network datasets. Our results reveal useful patterns, as well as the merits and pitfalls of each technique, paving the way for future work on this topic.

Cryptocurrencies are virtual currencies that exploit cryptography to perform secure financial transactions. They gained widespread popularity in recent years due to their decentralized nature, (pseudo-)anonymity, and ability to facilitate cross-border transactions without the need for intermediaries. However, their price on the market exhibits a huge volatility, that makes them prone to market anomalies. Therefore, predicting anomalies in cryptocurrency time series can be considered an important task for financial institutions, traders, and investors, to maximize their profit or minimize losses. In this paper, we propose a novel approach for predicting anomalies in cryptocurrency time series by exploiting temporal correlations among different cryptocurrencies. Our approach, called CARROT, is based on the idea that groups of cryptocurrencies exhibit similar trends, possibly due to common influencing factors. CARROT analyzes the temporal correlation between different cryptocurrencies, and identifies clusters showing similar patterns that can be useful for gaining insights into future anomalies. Subsequently, CARROT exploits multiple (i.e., one for each cluster) multi-target LSTM models to predict anomalies. Our experiments, performed on a dataset of 17 cryptocurrencies, proved that CARROT outperforms single-target LSTM models of up to 20%, as well as other approaches based on neural networks, i.e., MLP and CNN, in terms of macro F1-score. Therefore, the proposed approach can be considered as a promising tool for predicting anomalies in cryptocurrency time series data and can potentially be used to improve risk management and trading strategies in the cryptocurrency market. • Analysis of cryptocurrency trends. • Clustering-based multi-target prediction of anomalies in time series. • Consistent improvements achieved over the single-target counterpart.

The analysis of log data, generated by running processes in many application domains, enables organizations to identify opportunities for operational improvements. For instance, in healthcare, analyzing patient treatment logs can optimize care pathways; in manufacturing, production line logs can reveal bottlenecks; and in customer service, ticket resolution logs can streamline response protocols. One key analytical task is predicting the next activity in a process, which supports operational decision-making through better resource allocation and proactive response to customer needs. In this paper, we solve the next activity prediction task by exploiting a novel positional encoding approach that is based on sliding windows. This approach allows us to consider both a way to adapt to changes in the data distribution, and exploit positional information of the activities in the traces. The method proposed in this paper, called OREO, takes into account these aspects through a positional encoding tightly coupled with specific types of deep neural network architectures. The results on eight real-world process logs show the superiority of the models exploiting OREO encoding over state-of-the-art approaches, confirming our initial intuition of benefits gained by combining a time-window based model with positional information.

Abstract The rising adoption of AI models in real-world applications characterized by sensor data creates an urgent need for inference explanation mechanisms to support domain experts in making informed decisions. Explainable AI (XAI) opens up a new opportunity to extend black-box deep learning models with such inference explanation capabilities. However, existing XAI approaches for tabular, image, and graph data are ineffective in contexts with spatio-temporal data. In this paper, we fill this gap by proposing a XAI method specifically tailored for spatio-temporal data in sensor networks, where observations are collected at regular time intervals and at different locations. Our model-agnostic masking meta-optimization method for deep learning models uncovers global salient factors influencing model predictions, and generates explanations taking into account multiple analytical views, such as features, timesteps, and node locations. Our qualitative and quantitative experiments with real-world forecasting datasets show that our approach effectively extracts explanations of model predictions, and is competitive with state-of-the-art approaches.

Abstract Forecasting future measurements from geographically distributed sensors is essential across many application domains. However, the spatial distribution of these sensors raises multiple challenges, primarily due to spatial autocorrelation phenomena, that introduce inter-dependencies among nearby locations, that cannot therefore be treated independently by learning algorithms. While some existing approaches can capture such phenomena, they generally model the spatial dimension globally across all locations. On the other hand, the method we propose in this paper, called SPALT, focuses on capturing spatial relationships specifically among time series with similar trends, even if these trends occur at different times, thus modeling the spatio-temporal locality. SPALT leverages linear model trees, which allow us to naturally consider the spatial autocorrelation in a local manner: during the tree-building process, the adopted heuristics aim to group time series exhibiting similar trends into the same node, on which additional features considering the spatial dimension are selectively injected. Additionally, we propose a new pruning strategy, based on Reduced Error Pruning (REP), that also considers the spatio-temporal locality during the tree simplification. Designed for a multi-step setting, SPALT provides forecasts for multiple future time steps across multiple sensors simultaneously. The characteristics exhibited by SPALT can provide significant benefits in different domains, where measurements come from geographically distributed sensors. In this paper, we focus on data produced by sensors located in multiple renewable power plants measuring their energy production at regular, short intervals. Experiments on three real-world datasets demonstrate the effectiveness of SPALT in forecasting the production of energy at different time horizons, and its superior performance in comparison with tree-based models and state-of-the-art neural networks that incorporate both temporal and spatial dimensions.

In the era of Big Data, forecasting the future measurements of geo-distributed sensors can be considered among the most fundamental tasks for several application domains. However, the spatial distribution of such sensors introduces several challenges for forecasting methods. The most important one is due to the spatial dimension which introduces autocorrelation phenomena, according to which nearby locations are not independent and should not be treated as such by the learning algorithms. Some existing approaches are able to capture the spatial autocorrelation, but they tend to model spatial information globally across all locations. On the contrary, our method, called SPLiT, focuses on capturing spatial information only among time series with similar trends, also at different timing, modeling the so-called spatio-temporal locality. The proposed method is based on linear model trees, which allow us to naturally model any form of discontinuities in the spatial autocorrelation during the tree growing phase, guided by heuristics that identify time series with similar trends. SPLiT works in the multi-step predictive setting in order to simultaneously provide forecasts for multiple time steps for multiple sensors in the future.Our experiments, conducted on two real-world datasets, show the effectiveness of the proposed method in forecasting the energy production of geographically distributed renewable power plants. The comparison against several tree-based models and state-of-the-art neural networks that consider both temporal and spatial dimensions shows the superiority of the proposed method.

BACKGROUND: Microbiome dysbiosis has recently been associated with different diseases and disorders. In this context, machine learning (ML) approaches can be useful either to identify new patterns or learn predictive models. However, data to be fed to ML methods can be subject to different sampling, sequencing and preprocessing techniques. Each different choice in the pipeline can lead to a different view (i.e., feature set) of the same individuals, that classical (single-view) ML approaches may fail to simultaneously consider. Moreover, some views may be incomplete, i.e., some individuals may be missing in some views, possibly due to the absence of some measurements or to the fact that some features are not available/applicable for all the individuals. Multi-view learning methods can represent a possible solution to consider multiple feature sets for the same individuals, but most existing multi-view learning methods are limited to binary classification tasks or cannot work with incomplete views. RESULTS: We propose irBoost.SH, an extension of the multi-view boosting algorithm rBoost.SH, based on multi-armed bandits. irBoost.SH solves multi-class classification tasks and can analyze incomplete views. At each iteration, it identifies one winning view using adversarial multi-armed bandits and uses its predictions to update a shared instance weight distribution in a learning process based on boosting. In our experiments, performed on 5 multi-view microbiome datasets, the model learned by irBoost.SH always outperforms the best model learned from a single view, its closest competitor rBoost.SH, and the model learned by a multi-view approach based on feature concatenation, reaching an improvement of 11.8% of the F1-score in the prediction of the Autism Spectrum disorder and of 114% in the prediction of the Colorectal Cancer disease. CONCLUSIONS: The proposed method irBoost.SH exhibited outstanding performances in our experiments, also compared to competitor approaches. The obtained results confirm that irBoost.SH can fruitfully be adopted for the analysis of microbiome data, due to its capability to simultaneously exploit multiple feature sets obtained through different sequencing and preprocessing pipelines.

Semi-supervised learning (SSL) is a common approach to learning predictive models using not only labeled, but also unlabeled examples. While SSL for the simple tasks of classification and regression has received much attention from the research community, this is not the case for complex prediction tasks with structurally dependent variables, such as multi-label classification and hierarchical multi-label classification. These tasks may require additional information, possibly coming from the underlying distribution in the descriptive space provided by unlabeled examples, to better face the challenging task of simultaneously predicting multiple class labels. In this paper, we investigate this aspect and propose a (hierarchical) multi-label classification method based on semi-supervised learning of predictive clustering trees, which we also extend towards ensemble learning. Extensive experimental evaluation conducted on 24 datasets shows significant advantages of the proposed method and its extension with respect to their supervised counterparts. Moreover, the method preserves interpretability of classical tree-based models.

Abstract Dynamic networks are ubiquitous in many domains for modelling evolving graph-structured data and detecting changes allows us to understand the dynamic of the domain represented. A category of computational solutions is represented by the pattern-based change detectors (PBCDs), which are non-parametric unsupervised change detection methods based on observed changes in sets of frequent patterns over time. Patterns have the ability to depict the structural information of the sub-graphs, becoming a useful tool in the interpretation of the changes. Existing PBCDs often rely on exhaustive mining, which corresponds to the worst-case exponential time complexity, making this category of algorithms inefficient in practice. In fact, in such a case, the pattern mining process is even more time-consuming and inefficient due to the combinatorial explosion of the sub-graph pattern space caused by the inherent complexity of the graph structure. Non-exhaustive search strategies can represent a possible approach to this problem, also because not all the possible frequent patterns contribute to changes in the time-evolving data. In this paper, we investigate the viability of different heuristic approaches which prevent the complete exploration of the search space, by returning a concise set of sub-graph patterns (compared to the exhaustive case). The heuristics differ on the criterion used to select representative patterns. The results obtained on real-world and synthetic dynamic networks show that these solutions are effective, when mining patterns, and even more accurate when detecting changes.

Abstract The identification of anomalous activities is a challenging and crucially important task in sensor networks. This task is becoming increasingly complex with the increasing volume of data generated in real-world domains, and greatly benefits from the use of predictive models to identify anomalies in real time. A key use case for this task is the identification of misbehavior that may be caused by involuntary faults or deliberate actions. However, currently adopted anomaly detection methods are often affected by limitations such as the inability to analyze large-scale data, a reduced effectiveness when data presents multiple densities, a strong dependence on user-defined threshold configurations, and a lack of explainability in the extracted predictions. In this paper, we propose a distributed deep learning method that extends growing hierarchical self-organizing maps, originally designed for clustering tasks, to address anomaly detection tasks. The SOM-based modeling capabilities of the method enable the analysis of data with multiple densities, by exploiting multiple SOMs organized as a hierarchy. Our map-reduce implementation under Apache Spark allows the method to process and analyze large-scale sensor network data. An automatic threshold-tuning strategy reduces user efforts and increases the robustness of the method with respect to noisy instances. Moreover, an explainability component resorting to instance-based feature ranking emphasizes the most salient features influencing the decisions of the anomaly detection model, supporting users in their understanding of raised alerts. Experiments are conducted on five real-world sensor network datasets, including wind and photovoltaic energy production, vehicular traffic, and pedestrian flows. Our results show that the proposed method outperforms state-of-the-art anomaly detection competitors. Furthermore, a scalability analysis reveals that the method is able to scale linearly as the data volume presented increases, leveraging multiple worker nodes in a distributed computing setting. Qualitative analyses on the level of anomalous pollen in the air further emphasize the effectiveness of our proposed method, and its potential in determining the level of danger in raised alerts.

Forecasting methods are important decision support tools in geo-distributed sensor networks. However, challenges such as the multivariate nature of data, the existence of multiple nodes, and the presence of spatio-temporal autocorrelation increase the complexity of the task. Existing forecasting methods are unable to address these challenges in a combined manner, resulting in a suboptimal model accuracy. In this article, we propose GAP-LSTM, a novel geo-distributed forecasting method that leverages the synergic interaction of graph convolution, attention-based long short-term memory (LSTM), 2-D-convolution, and latent memory states to effectively exploit spatio-temporal autocorrelation in multivariate data generated by multiple nodes, resulting in improved modeling capabilities. Our extensive evaluation, involving real-world datasets on traffic, energy, and pollution domains, showcases the ability of our method to outperform state-of-the-art forecasting methods. An ablation study confirms that all method components provide a positive contribution to the accuracy of the extracted forecasts. The method also provides an interpretable visualization that complements forecasts with additional insights for domain experts.

Abstract The massive adoption of social networks increased the need to analyze users’ data and interactions to detect and block the spread of propaganda and harassment behaviors, as well as to prevent actions influencing people towards illegal or immoral activities. In this paper, we propose HURI, a method for social network analysis that accurately classifies users assafeorrisky, according to their behavior in the social network. Specifically, the proposed hybrid approach leverages both the topology of the network of interactions and the semantics of the content shared by users, leading to an accurate classification also in the presence of noisy data, such as users who may appear to be risky due to the topic of their posts, but are actually safe according to their relationships. The strength of the proposed approach relies on the full and simultaneous exploitation of both aspects, giving each of them equal consideration during the combination phase. This characteristic makes HURI different from other approaches that fully consider only a single aspect and graft partial or superficial elements of the other into the first. The achieved performance in the analysis of a real-world Twitter dataset shows that the proposed method offers competitive performance with respect to eight state-of-the-art approaches.

Forecasting in real-world domains characterized by large-scale sensor data is a timely and challenging problem. Despite the widespread adoption of big data frameworks and platforms, the landscape of available scalable forecasting methods is restricted to few options, due to the inherent challenges for model training distribution that arise with more complex models, such as graph-based deep neural networks. In this paper, we propose a distributed forecasting approach for sensor networks based on a Graph Convolutional Long Short-Term Memory (GCN-LSTM) model architecture. Sub-models for each node of the sensor network are built separately, and trained independently on multiple worker nodes in a cluster environment, while preserving the spatio-temporal contribution provided by the global structure of the network. Job distribution and scheduling are carried out through a custom strategy leveraging the Horovod framework, performing synchronous training on GPUs available on multiple worker nodes. Experiments on multi-node renewable energy datasets show that our approach is effective in terms of both accuracy and scalability. Specifically, the execution of the distributed workflow does not negatively impact the forecasting performance of the model, and the speedup and scaleup achieved are close to the ideal values.

Anomaly detection is a machine learning task that has been investigated within diverse research areas and application domains. In this paper, we performed anomaly detection for air pollution and public transport traffic analysis for the city of Oslo, Norway. To this aim, the state-of-the-art method SparkGHSOM was considered to learn predictive models for normal (i.e. regular) scenarios of air quality and traffic jams in a distributed fashion. Furthermore, we extended the main algorithm to make the detected anomalies explainable through an instance-based feature ranking approach. The results showed that SparkGHSOM is able to detect anomalies for both the real applications considered in this study, despite the fact it was designed for different tasks.

Transfer learning focuses on enhancing predictive models for a target domain, by exploiting the knowledge coming from a related source domain. However, most existing transfer learning methods assume that source and target domains are described with the same feature spaces. Heterogeneous transfer learning approaches aim to overcome this limitation, but they usually introduce strong assumptions (e.g., on the number of features), cannot distribute the workload to handle large volumes of data, or cannot work in challenging settings like the Positive-Unlabeled (PU) setting, where only positive and unlabelled examples are available. In this paper, we present a novel heterogeneous distributed transfer learning method that can work also in PU learning setting and overcomes all such limitations.The experimental evaluation was conducted in the context of a link prediction task in the biological domain. The results showed the effectiveness of the proposed method, that outperformed three state-of-the-art heterogeneous transfer learning approaches.

MOTIVATION: Gene regulation is responsible for controlling numerous physiological functions and dynamically responding to environmental fluctuations. Reconstructing the human network of gene regulatory interactions is thus paramount to understanding the cell functional organization across cell types, as well as to elucidating pathogenic processes and identifying molecular drug targets. Although significant effort has been devoted towards this direction, existing computational methods mainly rely on gene expression levels, possibly ignoring the information conveyed by mechanistic biochemical knowledge. Moreover, except for a few recent attempts, most of the existing approaches only consider the information of the organism under analysis, without exploiting the information of related model organisms. RESULTS: We propose a novel method for the reconstruction of the human gene regulatory network, based on a transfer learning strategy that synergically exploits information from human and mouse, conveyed by gene-related metabolic features generated in silico from gene expression data. Specifically, we learn a predictive model from metabolic activity inferred via tissue-specific metabolic modelling of artificial gene knockouts. Our experiments show that the combination of our transfer learning approach with the constructed metabolic features provides a significant advantage in terms of reconstruction accuracy, as well as additional clues on the contribution of each constructed metabolic feature. AVAILABILITY AND IMPLEMENTATION: The method, the datasets and all the results obtained in this study are available at: https://doi.org/10.6084/m9.figshare.c.5237687. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

As the complexity of data increases, so does the importance of powerful representations, such as relational and logical representations, as well as the need for machine learning methods that can learn predictive models in such representations. A characteristic of these representations is that they give rise to a huge number of features to be considered, thus drastically increasing the difficulty of learning in terms of computational complexity and the curse of dimensionality. Despite this, methods for ranking features in this context, i.e., estimating their importance are practically non-existent. Among the most well-known methods for feature ranking are those based on ensembles, and in particular tree ensembles. To develop methods for feature ranking in a relational context, we adopt the relational tree ensemble approach. We thus first develop methods for learning ensembles of relational trees, extending a wide spectrum of tree-based ensemble methods from the propositional to the relational context, resulting in methods for bagging and random forests of relational trees, as well as gradient boosted ensembles thereof. Complex relational features are considered in our ensembles: by using complex aggregates, we extend the standard collection of features that correspond to existential queries, such as ‘Does this person have any children?’, to more complex features that correspond to aggregation queries, such as ‘What is the average age of this person’s children?’. We also calculate feature importance scores and rankings from the different kinds of relational tree ensembles learned, with different kinds of relational features. The rankings provide insight into and explain the ensemble models, which would be otherwise difficult to understand. We compare the methods for learning single trees and different tree ensembles, using only existential qualifiers and using the whole set of relational features, against 10 state-of-the-art methods on a collection of benchmark relational datasets, deriving also the corresponding feature rankings. Overall, the bagging ensembles perform the best, with gradient boosted ensembles following closely. The use of aggregates is beneficial and in some datasets drastically improves performance: In these cases, aggregate-based features clearly stand out in the feature rankings derived from the ensembles.

Abstract Matrix tri-factorization subject to binary constraints is a versatile and powerful framework for the simultaneous clustering of observations and features, also known as biclustering. Applications for biclustering encompass the clustering of high-dimensional data and explorative data mining, where the selection of the most important features is relevant. Unfortunately, due to the lack of suitable methods for the optimization subject to binary constraints, the powerful framework of biclustering is typically constrained to clusterings which partition the set of observations or features. As a result, overlap between clusters cannot be modelled and every item, even outliers in the data, have to be assigned to exactly one cluster. In this paper we propose Broccoli , an optimization scheme for matrix factorization subject to binary constraints, which is based on the theoretically well-founded optimization scheme of proximal stochastic gradient descent. Thereby, we do not impose any restrictions on the obtained clusters. Our experimental evaluation, performed on both synthetic and real-world data, and against 6 competitor algorithms, show reliable and competitive performance, even in presence of a high amount of noise in the data. Moreover, a qualitative analysis of the identified clusters shows that Broccoli may provide meaningful and interpretable clustering structures.

Abstract A sitemap represents an explicit specification of the design concept and knowledge organization of a website and is therefore considered as the website’s basic ontology. It not only presents the main usage flows for users, but also hierarchically organizes concepts of the website. Typically, sitemaps are defined by webmasters in the very early stages of the website design. However, during their life websites significantly change their structure, their content and their possible navigation paths. Even if this is not the case, webmasters can fail to either define sitemaps that reflect the actual website content or, vice versa, to define the actual organization of pages and links which do not reflect the intended organization of the content coded in the sitemaps. In this paper we propose an approach which automatically generates sitemaps. Contrary to other approaches proposed in the literature, which mainly generate sitemaps from the textual content of the pages, in this work sitemaps are generated by analyzing the Web graph of a website. This allows us to: i) automatically generate a sitemap on the basis of possible navigation paths, ii) compare the generated sitemaps with either the sitemap provided by the Web designer or with the intended sitemap of the website and, consequently, iii) plan possible website re-organization. The solution we propose is based on closed frequent sequence extraction and only concentrates on hyperlinks organized in “Web lists”, which are logical lists embedded in the pages. These “Web lists” are typically used for supporting users in Web site navigation and they include menus, navbars and content tables. Experiments performed on three real datasets show that the extracted sitemaps are much more similar to those defined by website curators than those obtained by competitor algorithms.

MOTIVATION: The reconstruction of gene regulatory networks (GRNs) from gene expression data has received increasing attention in recent years, due to its usefulness in the understanding of regulatory mechanisms involved in human diseases. Most of the existing methods reconstruct the network through machine learning approaches, by analyzing known examples of interactions. However, (i) they often produce poor results when the amount of labeled examples is limited, or when no negative example is available and (ii) they are not able to exploit information extracted from GRNs of other (better studied) related organisms, when this information is available. RESULTS: In this paper, we propose a novel machine learning method that overcomes these limitations, by exploiting the knowledge about the GRN of a source organism for the reconstruction of the GRN of the target organism, by means of a novel transfer learning technique. Moreover, the proposed method is natively able to work in the positive-unlabeled setting, where no negative example is available, by fruitfully exploiting a (possibly large) set of unlabeled examples. In our experiments, we reconstructed the human GRN, by exploiting the knowledge of the GRN of Mus musculus. Results showed that the proposed method outperforms state-of-the-art approaches and identifies previously unknown functional relationships among the analyzed genes. AVAILABILITY AND IMPLEMENTATION: http://www.di.uniba.it/∼mignone/systems/biosfer/index.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: The study of functional associations between ncRNAs and human diseases is a pivotal task of modern research to develop new and more effective therapeutic approaches. Nevertheless, it is not a trivial task since it involves entities of different types, such as microRNAs, lncRNAs or target genes whose expression also depends on endogenous or exogenous factors. Such a complexity can be faced by representing the involved biological entities and their relationships as a network and by exploiting network-based computational approaches able to identify new associations. However, existing methods are limited to homogeneous networks (i.e., consisting of only one type of objects and relationships) or can exploit only a small subset of the features of biological entities, such as the presence of a particular binding domain, enzymatic properties or their involvement in specific diseases. RESULTS: To overcome the limitations of existing approaches, we propose the system LP-HCLUS, which exploits a multi-type hierarchical clustering method to predict possibly unknown ncRNA-disease relationships. In particular, LP-HCLUS analyzes heterogeneous networks consisting of several types of objects and relationships, each possibly described by a set of features, and extracts multi-type clusters that are subsequently exploited to predict new ncRNA-disease associations. The extracted clusters are overlapping, hierarchically organized, involve entities of different types, and allow LP-HCLUS to catch multiple roles of ncRNAs in diseases at different levels of granularity. Our experimental evaluation, performed on heterogeneous attributed networks consisting of microRNAs, lncRNAs, diseases, genes and their known relationships, shows that LP-HCLUS is able to obtain better results with respect to existing approaches. The biological relevance of the obtained results was evaluated according to both quantitative (i.e., TPR@k, Areas Under the TPR@k, ROC and Precision-Recall curves) and qualitative (i.e., according to the consultation of the existing literature) criteria. CONCLUSIONS: The obtained results prove the utility of LP-HCLUS to conduct robust predictive studies on the biological role of ncRNAs in human diseases. The produced predictions can therefore be reliably considered as new, previously unknown, relationships among ncRNAs and diseases.

In IoT environments, data are collected by many distinct devices, at the periphery, so that their feature-sets can be naturally endowed with a faceted structure. In this work, we argue that the IoT requires specialized ML models, able to exploit this faceted structure in the learning strategy. We demonstrate the application of this principle, by a multiple kernel learning approach, based on the exploration of the partition lattice driven by the natural partitioning of the feature set. Furthermore, we consider that the whole data management, acquisition, pre-processing and analytics pipeline results from the composition of processes pursuing different and non-perfectly aligned goals (most often, enacted by distinct agents with different constraints, requirements competencies and with non-aligned interests). We propose the adoption of an adversarial modeling paradigm across the overall pipeline. We argue that knowledge of the composite nature of the learning process, as well as of the adversarial character of the relationship among phases, can help in developing heuristics for improving the learning algorithms efficiency and accuracy. We develop our argument with reference to few exemplary use cases.

Process mining is a research discipline that aims to discover, monitor and improve real processing using event logs. In this paper we tackle the problem of next activity prediction/recommendation via "nested prediction model" learning, that is, we first identify recurrent and frequent sequences of activities and then we learn a prediction model for each frequent sequence. The key principle underlying the design of the proposed solution is in the ability to process massive logs by means of a parallel and distributed solution (by exploiting the Spark parallel computation framework) which can make reasonable decisions in the absence of perfect models. Indeed, given the classical threshold for minimum support and a user-specified error bound, our approach exploits the Chernoff bound to mine "approximate" frequent sequences with statistical error guarantees on their actual supports. Experiments on real-world log data prove the effectiveness of the proposed approach.

Heterogeneous networks are networks consisting of different types of objects and links. They can be found in several fields, ranging from the Internet to social sciences, biology, epidemiology, geography, finance, and many others. In the literature, several methods have been proposed for the analysis of network data, but they usually focus on homogeneous networks, where all the objects are of the same type, and links among them describe a single type of relationship. More recently, the complexity of real scenarios has impelled researchers to design methods for the analysis of heterogeneous networks, especially focused on classification and clustering tasks. However, they often make assumptions on the structure of the network that are too restrictive or do not fully exploit different forms of network correlation and autocorrelation. Moreover, when nodes which are the main subject of the classification task are linked to several nodes of the network having missing values, standard methods can lead to either building incomplete classification models or to discarding possibly relevant dependencies (correlation or autocorrelation). In this paper, we propose an ensemble learning approach for multi-type classification. We adopt the system Mr-SBC, which is originally able to analyze heterogeneous networks of arbitrary structure, within an ensemble learning approach. The ensemble allows us to improve the classification accuracy of Mr-SBC by exploiting i) the possible presence of correlation and autocorrelation phenomena and ii) the classification of instances (which contain missing values) of other node types in the network. As a beneficial side effect, we have also that the models are more stable in terms of standard deviation of the accuracy, over different samples used for training. Experiments performed on real-world datasets show that the proposed method is able to significantly outperform the standard implementation of Mr-SBC. Moreover, it gives Mr-SBC the advantage of outperforming four other well-known algorithms for the classification of data organized in a network.

In this paper, we tackle the problem of power prediction of several photovoltaic (PV) plants spread over an extended geographic area and connected to a power grid. The paper is intended to be a comprehensive study of one-day ahead forecast of PV energy production along several dimensions of analysis: 1) The consideration of the spatio-temporal autocorrelation, which characterizes geophysical phenomena, to obtain more accurate predictions. 2) The learning setting to be considered, i.e., using simple output prediction for each hour or structured output prediction for each day. 3) The learning algorithms: We compare artificial neural networks, most often used for PV prediction forecast, and regression trees for learning adaptive models. The results obtained on two PV power plant datasets show that: taking into account spatio/temporal autocorrelation is beneficial; the structured output prediction setting significantly outperforms the nonstructured output prediction setting; and regression trees provide better models than artificial neural networks.

Predicting the output power of renewable energy production plants distributed on a wide territory is a really valuable goal, both for marketing and energy management purposes. Vi-POC (Virtual Power Operating Center) project aims at designing and implementing a prototype which is able to achieve this goal. Due to the heterogeneity and the high volume of data, it is necessary to exploit suitable Big Data analysis techniques in order to perform a quick and secure access to data that cannot be obtained with traditional approaches for data management. In this paper, we describe Vi-POC -- a distributed system for storing huge amounts of data, gathered from energy production plants and weather prediction services. We use HBase over Hadoop framework on a cluster of commodity servers in order to provide a system that can be used as a basis for running machine learning algorithms. Indeed, we perform one-day ahead forecast of PV energy production based on Artificial Neural Networks in two learning settings, that is, structured and non-structured output prediction. Preliminary experimental results confirm the validity of the approach, also when compared with a baseline approach.

BACKGROUND: The understanding of mechanisms and functions of microRNAs (miRNAs) is fundamental for the study of many biological processes and for the elucidation of the pathogenesis of many human diseases. Technological advances represented by high-throughput technologies, such as microarray and next-generation sequencing, have significantly aided miRNA research in the last decade. Nevertheless, the identification of true miRNA targets and the complete elucidation of the rules governing their functional targeting remain nebulous. Computational tools have been proven to be fundamental for guiding experimental validations for the discovery of new miRNAs, for the identification of their targets and for the elucidation of their regulatory mechanisms. DESCRIPTION: ComiRNet (Co-clustered miRNA Regulatory Networks) is a web-based database specifically designed to provide biologists and clinicians with user-friendly and effective tools for the study of miRNA-gene target interaction data and for the discovery of miRNA functions and mechanisms. Data in ComiRNet are produced by a combined computational approach based on: 1) a semi-supervised ensemble-based classifier, which learns to combine miRNA-gene target interactions (MTIs) from several prediction algorithms, and 2) the biclustering algorithm HOCCLUS2, which exploits the large set of produced predictions, with the associated probabilities, to identify overlapping and hierarchically organized biclusters that represent miRNA-gene regulatory networks (MGRNs). CONCLUSIONS: ComiRNet represents a valuable resource for elucidating the miRNAs' role in complex biological processes by exploiting data on their putative function in the context of MGRNs. ComiRnet currently stores about 5 million predicted MTIs between 934 human miRNAs and 30,875 mRNAs, as well as 15 bicluster hierarchies, each of which represents MGRNs at different levels of granularity. The database can be freely accessed at: http://comirnet.di.uniba.it.

The problem of accurately predicting the energy production from renewable sources has recently received an increasing attention from both the industrial and the research communities. It presents several challenges, such as facing with the rate data are provided by sensors, the heterogeneity of the data collected, power plants efficiency, as well as uncontrollable factors, such as weather conditions and user consumption profiles. In this paper we describe Vi-POC (Virtual Power Operating Center), a project conceived to assist energy producers and decision makers in the energy market. In this paper we present the Vi-POC project and how we face with challenges posed by the specific application. The solutions we propose have roots both in big data management and in stream data mining.

BACKGROUND: MicroRNAs (miRNAs) are small non-coding RNAs which play a key role in the post-transcriptional regulation of many genes. Elucidating miRNA-regulated gene networks is crucial for the understanding of mechanisms and functions of miRNAs in many biological processes, such as cell proliferation, development, differentiation and cell homeostasis, as well as in many types of human tumors. To this aim, we have recently presented the biclustering method HOCCLUS2, for the discovery of miRNA regulatory networks. Experiments on predicted interactions revealed that the statistical and biological consistency of the obtained networks is negatively affected by the poor reliability of the output of miRNA target prediction algorithms. Recently, some learning approaches have been proposed to learn to combine the outputs of distinct prediction algorithms and improve their accuracy. However, the application of classical supervised learning algorithms presents two challenges: i) the presence of only positive examples in datasets of experimentally verified interactions and ii) unbalanced number of labeled and unlabeled examples. RESULTS: We present a learning algorithm that learns to combine the score returned by several prediction algorithms, by exploiting information conveyed by (only positively labeled/) validated and unlabeled examples of interactions. To face the two related challenges, we resort to a semi-supervised ensemble learning setting. Results obtained using miRTarBase as the set of labeled (positive) interactions and mirDIP as the set of unlabeled interactions show a significant improvement, over competitive approaches, in the quality of the predictions. This solution also improves the effectiveness of HOCCLUS2 in discovering biologically realistic miRNA:mRNA regulatory networks from large-scale prediction data. Using the miR-17-92 gene cluster family as a reference system and comparing results with previous experiments, we find a large increase in the number of significantly enriched biclusters in pathways, consistent with miR-17-92 functions. CONCLUSION: The proposed approach proves to be fundamental for the computational discovery of miRNA regulatory networks from large-scale predictions. This paves the way to the systematic application of HOCCLUS2 for a comprehensive reconstruction of all the possible multiple interactions established by miRNAs in regulating the expression of gene networks, which would be otherwise impossible to reconstruct by considering only experimentally validated interactions.

Multi-Relational Data Mining (MRDM) refers to the process of discovering implicit, previously unknown and potentially useful information from data scattered in multiple tables of a relational database. Following the mainstream of MRDM research, we tackle the regression where the goal is to examine samples of past experience with known continuous answers (response) and generalize future cases through an inductive process. Mr-SMOTI, the solution we propose, resorts to the structural approach in order to recursively partition data stored into a tightly-coupled database and build a multi-relational model tree which captures the linear dependence between the response variable and one or more explanatory variables. The model tree is top-down induced by choosing, at each step, either to partition the training space or to introduce a regression variable in the linear models with the leaves. The tight-coupling with the database makes the knowledge on data structures (foreign keys) available free of charge to guide the search in the multi-relational pattern space. Experiments on artificial and real databases demonstrate that in general Mr-SMOTI outperforms both SMOTI and M5' which are two propositional model tree induction systems, and TILDE-RT which is a state-of-art structural model tree induction system.

BACKGROUND: microRNAs (miRNAs) are a class of small non-coding RNAs which have been recognized as ubiquitous post-transcriptional regulators. The analysis of interactions between different miRNAs and their target genes is necessary for the understanding of miRNAs' role in the control of cell life and death. In this paper we propose a novel data mining algorithm, called HOCCLUS2, specifically designed to bicluster miRNAs and target messenger RNAs (mRNAs) on the basis of their experimentally-verified and/or predicted interactions. Indeed, existing biclustering approaches, typically used to analyze gene expression data, fail when applied to miRNA:mRNA interactions since they usually do not extract possibly overlapping biclusters (miRNAs and their target genes may have multiple roles), extract a huge amount of biclusters (difficult to browse and rank on the basis of their importance) and work on similarities of feature values (do not limit the analysis to reliable interactions). RESULTS: To overcome these limitations, HOCCLUS2 i) extracts possibly overlapping biclusters, to catch multiple roles of both miRNAs and their target genes; ii) extracts hierarchically organized biclusters, to facilitate bicluster browsing and to distinguish between universe and pathway-specific miRNAs; iii) extracts highly cohesive biclusters, to consider only reliable interactions; iv) ranks biclusters according to the functional similarities, computed on the basis of Gene Ontology, to facilitate bicluster analysis. CONCLUSIONS: Our results show that HOCCLUS2 is a valid tool to support biologists in the identification of context-specific miRNAs regulatory modules and in the detection of possibly unknown miRNAs target genes. Indeed, results prove that HOCCLUS2 is able to extract cohesiveness-preserving biclusters, when compared with competitive approaches, and statistically confirm (at a confidence level of 99%) that mRNAs which belong to the same biclusters are, on average, more functionally similar than mRNAs which belong to different biclusters. Finally, the hierarchy of biclusters provides useful insights to understand the intrinsic hierarchical organization of miRNAs and their potential multiple interactions on target genes.

BACKGROUND: Ontologies and catalogs of gene functions, such as the Gene Ontology (GO) and MIPS-FUN, assume that functional classes are organized hierarchically, that is, general functions include more specific ones. This has recently motivated the development of several machine learning algorithms for gene function prediction that leverages on this hierarchical organization where instances may belong to multiple classes. In addition, it is possible to exploit relationships among examples, since it is plausible that related genes tend to share functional annotations. Although these relationships have been identified and extensively studied in the area of protein-protein interaction (PPI) networks, they have not received much attention in hierarchical and multi-class gene function prediction. Relations between genes introduce autocorrelation in functional annotations and violate the assumption that instances are independently and identically distributed (i.i.d.), which underlines most machine learning algorithms. Although the explicit consideration of these relations brings additional complexity to the learning process, we expect substantial benefits in predictive accuracy of learned classifiers. RESULTS: This article demonstrates the benefits (in terms of predictive accuracy) of considering autocorrelation in multi-class gene function prediction. We develop a tree-based algorithm for considering network autocorrelation in the setting of Hierarchical Multi-label Classification (HMC). We empirically evaluate the proposed algorithm, called NHMC (Network Hierarchical Multi-label Classification), on 12 yeast datasets using each of the MIPS-FUN and GO annotation schemes and exploiting 2 different PPI networks. The results clearly show that taking autocorrelation into account improves the predictive performance of the learned models for predicting gene function. CONCLUSIONS: Our newly developed method for HMC takes into account network information in the learning phase: When used for gene function prediction in the context of PPI networks, the explicit consideration of network autocorrelation increases the predictive performance of the learned models. Overall, we found that this holds for different gene features/ descriptions, functional annotation schemes, and PPI networks: Best results are achieved when the PPI network is dense and contains a large proportion of function-relevant interactions.

The automatic discovery of process models can help to gain insight into various perspectives (e.g., control flow or data perspective) of the process executions traced in an event log. Frequent patterns mining offers a means to build human understandable representations of these process models. This paper describes the application of a multi-relational method of frequent pattern discovery into process mining. Multi-relational data mining is demanded for the variety of activities and actors involved in the process executions traced in an event log which leads to a relational (or structural) representation of the process executions. Peculiarity of this work is in the integration of disjunctive forms into relational patterns discovered from event logs. The introduction of disjunctive forms enables relational patterns to express frequent variants of process models. The effectiveness of using relational patterns with disjunctions to describe process models with variants is assessed on real logs of process executions.

The amount of data produced by ubiquitous computing applications is quickly growing, due to the pervasive presence of small devices endowed with sensing, computing and communication capabilities. Heterogeneity and strong interdependence, which characterize ‘ubiquitous data’, require a (multi-)relat ional approach to their analysis. However, relational data mining algorithms do not scale well and very large data sets are hardly processable. In this paper we propose an extension of a relational algorithm for multi-level frequent pattern discovery, which resorts to data sampling and distributed computation in Grid environments, in order to overcome the computational limits of the original serial algorithm. The set of patterns discovered by the new algorithm approximates the set of exact solutions found by the serial algorithm. The quality of approximation depends on three parameters: the proportion of data in each sample, the minimum support thresholds and the number of samples in which a pattern has to be frequent in order to be considered globally frequent. Considering that the first two parameters are hardly controllable, we focus our investigation on the third one. Theoretically derived conclusions are also experimentally confirmed. Moreover, an additional application in the context of event log mining proves the viability of the proposed approach to relational frequent pattern mining from very large data sets.

In recent years, a growing interest has been given to trajectory data mining applications that permit to support mobility prediction with the aim of anticipating or pre-fetching possible services [3]. Proposed approaches typically consider only spatio-temporal information provided by collected trajectories. However, in some scenarios, such as that of tourist supporting, semantic information which express needs and interest of the user (tourist) should be taken into account. This semantic information can be extracted from textual documents already consulted by the tourists. In this paper, we present the application of the time-slice density estimation [1] that permits to suggest/predict the next destination of the tourist. In particular, time-slice density estimation permits to measure the rate of change of tourist’s interests at a given geographical position over a user-defined time horizon. Tourist interests depend both on the geographical position of the tourist with respect to a reference system and on semantic information provided by geo-referenced documents associated to the visited sites. Our basic assumption is that a tourist moves towards a close destination which is semantically consistent with her/his current profile as much as possible. Destinations that minimize the profile drift are suggested/predicted as next destinations.

Analyzing physiological data can be of great importance in unearthing information on the course of a disease. In this paper we propose a data mining approach to analyze these data and acquire knowledge, in the form of temporal patterns, on the physiological events which can frequently trigger particular stages of disease. The application to the sleep sickness scenario is addressed to discover patterns, expressed in terms of breathing and cardiovascular system time-annotated disorders, which may trigger particular sleep stages.

Determining the reading order for layout components ex- tracted from a document image can be a crucial problem for several applications. It enables the reconstruction of a single textual element from texts associated to multiple layout components and makes both information extraction and content-based retrieval of documents more effective. A common aspect for all methods reported in the literature is that they strongly depend on the specific domain and are scarcely reusable when the classes of documents or the task at hand changes. In this paper, we investigate the prob- lem of detecting the reading order of layout components by resorting to a data mining approach which acquires the do- main specific knowledge from a set of training examples. The input of the learning method is the description of the "chains" of layout components defined by the user. Only spatial information is exploited to describe a chain, thus making the proposed approach also applicable to the cases in which no text can be associated to a layout component. The method induces a probabilistic classifier based on the Bayesian framework which is used for reconstructing either single or multiple chains of layout components. It has been evaluated on a set of document images.

Processing censorship cards of the 20/sup th/ century in order to support annotation and retrieval processes, leads to a number of challenges for many DIA systems. Problems due to the low layout quality and standard of such a material can be reduced by exploiting information conveyed by color. In this paper, taking into account lessons learned in the context of the 1ST project Collate, we propose a new method for image segmentation and layout analysis that takes full advantage of color information. The method has been implemented in the DIA system WISDOM++ and tested on a corpus of multi-format documents concerning historic film censorships.

In this paper, we evaluate and systematically compare two different (multi-)relational learning methods based on a statistical approach and a logical approach for the task of document image understanding. For a fair comparison, both methods are tested on the same real world dataset consisting of multipage articles published in an international journal. An analysis of pros and cons of both approaches is reported.

We consider the problem of eliminating redundant Boolean features for a given data set, where a feature is redundant if it separates the classes less well than another feature or set of features. Lavrač et al. proposed the algorithm REDUCE that works by pairwise comparison of features, i.e., it eliminates a feature if it is redundant with respect to another feature. Their algorithm operates in an ILP setting and is restricted to two-class problems. In this paper we improve their method and extend it to multiple classes. Central to our approach is the notion of a neighbourhood of examples: a set of examples of the same class where the number of different features between examples is relatively small. Redundant features are eliminated by applying a revised version of the REDUCE method to each pair of neighbourhoods of different class. We analyse the performance of our method on a range of data sets.

Model trees are an extension of regression trees that associate leaves with multiple regression models. In this paper, a method for the data-driven construction of model trees is presented, namely, the Stepwise Model Tree Induction (SMOTI) method. Its main characteristic is the induction of trees with two types of nodes: regression nodes, which perform only straight-line regression, and splitting nodes, which partition the feature space. The multiple linear model associated with each leaf is then built stepwise by combining straight-line regressions reported along the path from the root to the leaf. In this way, internal regression nodes contribute to the definition of multiple models and have a "global" effect, while straight-line regressions at leaves have only "local" effects. Experimental results on artificially generated data sets show that SMOTI outperforms two model tree induction systems, M5' and RETIS, in accuracy. Results on benchmark data sets used for studies on both regression and model trees show that SMOTI performs better than RETIS in accuracy, while it is not possible to draw statistically significant conclusions on the comparison with M5'. Model trees induced by SMOTI are generally simple and easily interpretable and their analysis often reveals interesting patterns.

In this paper, a machine learning approach to support the user during the correction of the layout analysis is proposed. Layout analysis is the process of extracting a hierarchical structure describing the layout of a page. In our approach, the layout analysis is performed in two steps: firstly, the global analysis determines possible areas containing paragraphs, sections, columns, figures and tables, and secondly, the local analysis groups together blocks that possibly fall within the same area. The result of the local analysis process strongly depends on the quality of the results of the first step. We investigate the possibility of supporting the user during the correction of the results of the global analysis. This is done by allowing the user to correct the results of the global analysis and then by learning rules for layout correction from the sequence of user actions. Experimental results on a set of multi-page documents are reported and commented.