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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.