Predictive process monitoring (PPM) focuses on predicting future process trajectories, including next activity predictions. This is crucial in dynamic environments where processes change or face uncertainty. However, current frameworks often assume a static environment, overlooking dynamic characteristics and concept drifts. This results in catastrophic forgetting, where training while focusing merely on new data distribution negatively impacts the performance on previously learned data distributions. Continual learning addresses, among others, the challenges related to mitigating catastrophic forgetting. This paper proposes a novel approach called Continual Next Activity Prediction with Prompts (CNAPwP), which adapts the DualPrompt algorithm for next activity prediction to improve accuracy and adaptability while mitigating catastrophic forgetting. We introduce new datasets with recurring concept drifts, alongside a task-specific forgetting metric that measures the prediction accuracy gap between initial occurrence and subsequent task occurrences. Extensive testing on three synthetic and two real-world datasets representing several setups of recurrent drifts shows that CNAPwP achieves SOTA or competitive results compared to five baselines, demonstrating its potential applicability in real-world scenarios. An open-source implementation of our method, together with the datasets and results, is available at: https://github.com/SvStraten/CNAPwP .

Cities have been growing ever larger especially in the most recent decades. As neighborhoods and boundaries of cities are expanding, the resident-service related problems within these cities increase. Such problems are often reported by residents, and city administrators rely on the submission of these complaints for an accurate depiction of the most prevalent issues in a neighborhood. We address the problem of predicting when new citizen complaints are submitted and when an increase of complaints is likely to occur in a city. The submitted complaints are a type of spatiotemporal data, of which it was not well understood in earlier research how to use both the spatial and temporal dimension for complaint prediction. In this paper, a generalized framework is proposed that uses both the spatial and the temporal dimensions simultaneously to predict outliers of a (primary) stream of data using a possibly correlated secondary stream of data. It computes the optimal offset between the secondary stream in regard to the primary stream of data and trains a semi-supervised model to predict unseen data. Subsequently, outliers are detected from the predicted data. To evaluate the framework, we use a dataset that contain human mobility in NYC as one stream and another dataset with complaints as the other stream. The experimental evaluation showed promising results in an absolute sense and also compared to a baseline model when using the error in the predicted number of submitted complaints. Particularly for interesting complaints, the error between the prediction and ground truth is minimal. The tool and the datasets of this work are publicly available under https://github.com/Staartvin/outlier-prediction-framework

Process mining focuses on applying data mining techniques over business process data. Recently, with the improvements in sensoring, collection, and storage of event data, a big demand for both shorter mining time and adaptive models of streaming process events arose. This increased the interest in streaming process mining. Some techniques within this field attempt to identify drifts (change points) from evolving process data streams. Existing work on supervised and unsupervised-learning approaches over data streams have several limitations with regards to the nature of the drifts, the excessive storage required to store and process the stream, and the performance over real-world datasets. This paper contributes PrefixCDD, an efficient unsupervised-learning novel approach for online concept drift detection (CDD) over event streams. Our proposed approach utilizes a data structure, where the data stream components are stored in a set of prefix-trees. It transforms then the discrete data into continuous one using a Principal Component Analysis (PCA) approach over the trees. Then, ADWIN is used to focus on up-to-date information, making it appealing to work with the decaying mechanism logic behind our algorithm. Using six artificial and three real-life datasets, PrefixCDD outperforms state-of-the-art techniques in terms of detecting existing drifts of different natures, discovering them shortly after they appear, and the overall execution time.

Conformance checking (CC) techniques in process mining determine the conformity of cases, by means of their event sequences, with respect to a business process model. Online conformance checking (OCC) techniques perform such analysis for cases in event streams. Cases in streams may essentially not be concluded. Therefore, OCC techniques usually neglect the memory limitation and store all the observed cases whether seemingly concluded or unconcluded. Such indefinite storage of cases is inconsistent with the spirit of privacy regulations, such as GDPR, which advocate the retention of minimal data for a definite period of time. Catering to the aforementioned constraints, we propose two classes of novel approaches that partially or fully forget cases but can still properly estimate the conformance of their future events. All our proposed approaches bound the number of cases in memory and forget those in excess of the defined limit on the basis of prudent forgetting criteria. One class of these proposed approaches retains a meaningful summary of the forgotten events in order to resume the CC of their cases in the future, while the other class leverages classification for this purpose. We highlight the effectiveness of all our proposed approaches compared to a state of the art OCC technique lacking any forgetting mechanism through experiments using real-life as well as synthetic event data under a streaming setting. Our approaches substantially reduce the amount of data required to be retained while minimally impacting the accuracy of the conformance statistics.

The Euclidean distance is one of the most commonly used distance metrics. Several approximations have been pro-posed in the literature to reduce the complexity of this metric for high-dimensional or large datasets. In this paper, we propose BitBooster, an approximation to the Euclidean distance that can be efficiently computed using binary operations and which can also be applied to the Manhattan distance. The introduced approximation error is shown to be negligible when BitBooster is used for both convex- and density-based clustering. While obtaining clusters of almost the same quality as those obtained with the exact computation, we require only a fraction of the computation time. We demonstrate the superiority of our method to alternative approximations on 960 synthetic and 13 real-world datasets of varying sizes, dimensions and clusters.

Recent approaches have proven the effectiveness of local outlier factor-based outlier detection when applied over traffic flow probability distributions. However, these approaches used distance metrics based on the Bhattacharyya coefficient when calculating probability distribution similarity. Consequently, the limited expressiveness of the Bhattacharyya coefficient restricted the accuracy of the methods. The crucial deficiency of the Bhattacharyya distance metric is its inability to compare distributions with non-overlapping sample spaces over the domain of natural numbers. Traffic flow intensity varies greatly, which results in numerous non-overlapping sample spaces, rendering metrics based on the Bhattacharyya coefficient inappropriate. In this work, we address this issue by exploring alternative distance metrics and showing their applicability in a massive real-life traffic flow data set from 26 vital intersections in The Hague. The results on these data collected from 272 sensors for more than two years show various advantages of the Earth Mover's distance both in effectiveness and efficiency.

Accurate traffic flow prediction is an important tool to allow for more efficient use of traffic networks. Current traffic flow prediction algorithms such as LSTM RNNs can be very successful in predicting regular traffic flows, but often fail to accurately predict the more interesting irregularities in the traffic flows. We propose OE-LSTM (Outlier-Enriched LSTM), a novel framework for traffic flow prediction that focuses mainly on these irregular traffic flows. We consider the irregularities as outliers within each traffic flow stream and assume that for these traffic outliers to occur, a certain set of circumstances is present to cause these deviations from the regular traffic flow pattern. After detecting these outliers in traffic flow data, we measure the relation between them by performing a spatiotemporal correlation analysis. We use these correlations to trace the context of the outliers and determine as such which elements are interesting to include in our OE-LSTM prediction model. In addition to this, we also draw the foundations of a reference Cloud-based architecture for supporting big traffic data prediction based on OE-LSTM. The experimental results prove the effectiveness of the OE-LSTM framework in a real world traffic flow dataset. Both the dataset and the implementation of our framework are provided.

Customer journey analysis is rapidly increasing in popularity, as it is essential for companies to understand how their customers think and behave. Recent studies investigate how customers traverse their journeys and how they can be improved for the future. However, those researches only focus on improving the process for future customers by analyzing the historical data. This research focuses on helping the current customer immediately, by analyzing if it is possible to predict what the customer will do next and accordingly take proactive steps. We propose a model to predict the customer's next contact type (touch point). At first we will analyze the customer journey data by applying process mining techniques. We will use these insights then together with the historical data of accumulated customer journeys to train several classifiers. The winning of those classifiers, namely XGBoost, is used to perform a prediction on a customer's journey while the journey is still active. We show on three different real datasets coming from interactions between a telecommunication company and its customers that we always beat a baseline classifier thanks to our thorough pre-processing of the data.

Process performance analysis is an important subtask of process mining that aims at optimizing the discovered process models. In this paper we focus on improving process throughput by predicting congestions in the process execution (bottlenecks). We discuss an ongoing work on incorporating gradual and seasonal concept drift in this bottleneck prediction. In the field of process mining, we develop a method of predicting whether and which bottleneck will likely appear based on data known before a case starts. We introduce GRAHOF, a Gradual and Recurrent Adaptive Hoeffding Option Forest approach, which adapts to gradual and seasonal concept drifts when predicting bottlenecks of business processes in an online setting. We evaluate the parameters involved in GRAHOF using a synthetic event stream and a real-world event log.

GDPR (General Data Protection Regulation) is a new regulation of the European Union that superimposes strict privacy constraints on storing, accessing and processing user data, as a way to ensure that personal user data are not violated neither disclosed without an explicit consent. As a consequence, business processes that interact with large amounts of such data may easily cause GDPR violations, due to the typical complexity of such processes. Inspired by these considerations, this paper highlights the challenges and critical aspects associated with the GDPR compliance journey when opting for naïve straight-forward solutions. We propose a business-aware GDPR compliance journey using online process mining. Using several large log files generated based on a real scenario, we show that the proposed tool is both effective and efficient. As such, it proves to be a powerful concept for usage in incremental GDPR compliance environments.

Process mining is an emerging data mining task of gathering valuable knowledge out of the huge collections of business operation data. Despite its relatively young age, it has successfully provided many new insights into business workflows using established data mining techniques. Recently, with the huge improvements in the technologies of sensoring, collection and storing of data, a big demand for both shorter mining times and adaptive models of streaming process events arose. This initiated the field of stream process mining very recently. Drifts in the underlying concepts of the business processes are of a great interest for decision makers. One important advantage of stream process mining techniques over static ones is the ability to detect such drifts and to adapt its models accordingly. In this paper, we introduce an efficient approach that uses the collected information of an event stream miner to detect concept drifts. We use a dynamic window, which grows in size for stationary process behavior and shrinks for diverting data and thus indicating a concept drift. This adaptive window is used to build a model by focusing only on up-to-date information and discarding outdated items. Extensive experimental evaluations over real and synthetic log files show the ability of our algorithm to detect sudden drifts. We additionally show the effectiveness of our concept detection method in setting the pruning period of a recent stream mining algorithm.

In this paper we introduce a technique for the synthetic generation of memory references which behave as those generated by given running programs. Our approach is based on a novel Machine Learning algorithm we called Hierarchical Hidden/non Hidden Markov Model (HHnHMM). Short chunks of memory references from a running program are classified as Sequential, Periodic, Random, Jump or Other. Such execution classes are used to train an HHnHMM for that program. Trained HHnHMM are used as stochastic generators of memory reference addresses. In this way we can generate in real time memory reference streams of any length, which mimic the behavior of given programs without the need to store anything.

This paper proposes and experimentally assesses a machine learning approach for supporting the effective and efficient generation of synthetic memory reference traces for a wide range of application scenarios. The proposed approach makes a nice use of extended hierarchical Markov models.

This paper address the problem of temporal pattern mining from multiple data streams containing temporal events. Temporal events are considered as real world events aligned with comprehensive starting and ending timing information rather than simple integer timestamps. Predefined relations, such as "before" and "after", describe the heterogeneous relationships hidden in temporal data with limited diversity. In this work, the relationships among events are learned dynamically from the temporal information. Each event is treated as an object with a label and numerical attributes. An online-offline model is used as the primary structure for analyzing the evolving multiple streams. Different distance functions on temporal events and sequences can be applied depending on the application scenario. A prefix tree is introduced for a fast incremental pattern update.

Big Data Streams are very popular at now, as stirred-up by a plethora of modern applications such as sensor networks, scientific computing tools, Web intelligence, social network analysis and mining tools, and so forth. Here, the main research issue consists in how to effectively and efficiently extract useful knowledge from (streaming) big data, in order to support innovative big data analytics platforms. To this end, clustering analysis is a well-known tool for extracting knowledge from big data streams, as also confirmed by recent trends in active literature. A special applicative case is represented by so-called graph-shaped data (big) streams, which are produced by graph sources providing both structure-and content-oriented knowledge. On top of such sources, big graph analytics is a leading scientific area to be considered. At the convergence of these emerging topics, in this paper we provide the following contributions: (i) I-HASTREAM, a novel density-based hierarchical clustering algorithm for evolving big data streams that founds on it predecessor, namely HASTREAM, (ii) the architecture of a big graph analytics engine that embeds I-HASTREAM in its core layer.

Almost all activities observed in nowadays applications are correlated with a timing sequence. Users are mainly looking for interesting sequences out of such data. Sequential pattern mining algorithms aim at nding frequent sequences. Usually, the mined activities have timing durations that represent time intervals between their starting and ending points. Most sequential pattern mining approaches dealt with such activities as a single point event and thus lost many valuable information in the collected patterns. We present the PIVOTMiner, an ecient interval-based sequential pattern mining algorithm using a geometric representation of intervals. The interestingness level is not necessarily positively correlated with the frequency of the patterns. In many applications, users are seeking for rare patterns that considerably deviate from the majority. Simply delivering the bottom-k patterns does not guarantee their high outlierness (or deviation) from the frequent ones. We propose additionally the PIVOTRanker, the rst scalable algorithm for ranking rare interval-based sequential patterns based on their outlierness. Our experimental results on both synthetic and real-world datasets show that PIVOTMiner spends considerably less time than two state-of-the-art competitors, and that PIVOTRanker delivers a meaningful and useful ranking of rare patterns.

Almost all activities observed in nowadays applications are correlated with a timing sequence. Users are mainly looking for interesting sequences out of such data. Sequential pattern mining algorithms aim at finding frequent sequences. Usually, the mined activities have timing durations that represent time intervals between their starting and ending points. The majority of sequential pattern mining approaches dealt with such activities as a single point event and thus lost valuable information in the collected patterns. Recently, some approaches have carefully considered this interval-based nature of the events, but they have major limitations. They concentrate only on the order of events without taking the durations of the gaps between them into account and usually employ a binary representation to describe patterns. To resolve these problems, we propose the PIVOTMiner, an interval-based data mining algorithm using a geometric representation approach of intervals. Noisy events can be served with the geometric representation and a fuzzy set can be retrieved from the geometric patterns. PIVOTMiner can flexibly work on data presented as any number of not necessarily aligned interval sequences and in particular can utilize data presented as single interval sequence stream without the need to create samples. Our experimental results on both synthetic and real-world smart home datasets show that the information presented in our mined patterns are richer than those of most state-of-the-art algorithms while spending considerably smaller running times.

The analysis of the temporal evolution of dynamic networks is a key challenge for understanding complex processes hidden in graph structured data. Graph evolution rules capture such processes on the level of small subgraphs by describing frequently occurring structural changes within a network. Existing rule discovery methods make restrictive assumptions on the change processes present in networks. We propose EvoMine, a frequent graph evolution rule mining method that, for the first time, supports networks with edge insertions and deletions as well as node and edge relabelings. EvoMine defines embedding-based and event-based support as two novel measures to assess the frequency of rules. These measures are based on novel mappings from dynamic networks to databases of union graphs that retain all evolution information relevant for rule mining. Using these mappings the rule mining problem can be solved by frequent subgraph mining. We evaluate our approach and two baseline algorithms on several real datasets. To the best of our knowledge, this is the first empirical comparison of rule mining algorithmsfor dynamic networks.

One of the most fundamental challenges when accessing gestural patterns in 3D motion capture databases is the definition of spatiotemporal similarity. While distance-based similarity models such as the Gesture Matching Distance on gesture signatures are able to leverage the spatial and temporal characteristics of gestural patterns, their applicability to large 3D motion capture databases is limited due to their high computational complexity. To this end, we present a lower bound approximation of the Gesture Matching Distance that can be utilized in an optimal multi-step query processing architecture in order to provide efficient query processing. We investigate the performance in terms of accuracy and efficiency based on 3D motion capture databases and show that our approach is able to achieve an increase in efficiency of more than one order of magnitude with a negligible loss in accuracy.

Networks entail vulnerable and sensitive information that pose serious privacy threats. In this paper, we introduce, k-core attack, a new attack model which stems from the k-core decomposition principle. K-core attack undermines the privacy of some state-of-the-art techniques. We propose a novel structural anonymization technique called (k, δ)-Core Anonymity, which harnesses the k-core attack and structurally anonymizes small and large networks. In addition, although real-world social networks are massive in nature, most existing works focus on the anonymization of networks with less than one hundred thousand nodes. (k, δ)-Core Anonymity is tailored for massive networks. To the best of our knowledge, this is the first technique that provides empirical studies on structural network anonymization for massive networks. Using three real and two synthetic datasets, we demonstrate the effectiveness of our technique on small and large networks with up to 1.7 million nodes and 17.8 million edges. Our experiments reveal that our approach outperforms a state-of-the-art work in several aspects.

Clustering of high dimensional streaming data is an emerging field of research. A real life data stream imposes many challenges on the clustering task, as an endless amount of data arrives constantly. A lot of research has been done in the full space stream clustering. To handle the varying speeds of the data stream, "anytime" algorithms are proposed but so far only in full space stream clustering. However, data streams from many application domains contain abundance of dimensions; the clusters often exist only in specific subspaces (subset of dimensions) and do not show up in the full feature space. In this paper, the first algorithm that considers both the high dimensionality and the varying speeds of streaming data, is proposed. The algorithm, called SubClusTree, can flexibly adapt to the different stream speeds and makes the best use of available time to provide a high quality subspace clustering. The experimental results prove the effectiveness of our anytime subspace concept.

Context prediction is an emerging topic in the fields of data mining and information management which is both promising and challenging. Predicting the location of mobile objects was a frequently tackled subtask of mobile context prediction in recent researches. For scenarios of managing health information of mobile persons, the prediction of near future health status of persons is at least equally important to predicting their location. We introduce in this paper, to the best of our knowledge, a first method for predicting a next health context of mobile persons equipped with body sensors and a mobile device. The suggested Prefix Span-based method searches for sequential patterns within multiple streaming inputs from the body sensors as well as other contextual streams that influence the health context. We discuss additionally the implementation of our method in an energy aware mobile-server environment.