Predicting Co-movement patterns in mobility data

GeoInformatica https://doi.org/10.1007/s10707-022-00478-x Predicting Co‑movement patterns in mobility data Andreas Tritsarolis1 · Eva Chondrodima1 · Panagiotis Tampakis2 · Aggelos Pikrakis1 · Yannis Theodoridis1 Received: 23 November 2021 / Revised: 2 August 2022 / Accepted: 8 September 2022 © The Author(s) 2022 Abstract Predictive analytics over mobility data is of great importance since it can assist an analyst to predict events, such as collisions, encounters, traffic jams, etc. A typical example is anticipated location prediction, where the goal is to predict the future location of a moving object, given a look-ahead time. What is even more challenging is to be able to accurately predict collective behavioural patterns of movement, such as co-movement patterns as well as their course over time. In this paper, we address the problem of Online Prediction of Comovement Patterns. Furthermore, in order to be able to calculate the accuracy of our solution, we propose a co-movement pattern similarity measure, which facilitates the comparison between the predicted clusters and the actual ones. Finally, we calculate the clusters’ evolution through time (survive, split, etc.) and compare the cluster evolution predicted by our framework with the actual one. Our experimental study uses two real-world mobility datasets from the maritime and urban domain, respectively, and demonstrates the effectiveness of the proposed framework. Keywords Machine learning · Predictive analytics · Co-movement patterns · Trajectory prediction · Cluster evolution * Andreas Tritsarolis Eva Chondrodima Panagiotis Tampakis Aggelos Pikrakis Yannis Theodoridis 1 Data Science Lab., Department of Informatics, University of Piraeus, Piraeus, Greece 2 Department of Mathematics and Computer Science, University of Southern Denmark, Odense, Denmark 13 Vol.:(0123456789) GeoInformatica 1 Introduction The vast spread of GPS-enabled devices, such as smartphones, tablets, and GPS trackers, has led to the production of large amounts of mobility related data. By nature, this kind of data is streaming and there are several application scenarios where the processing needs to take place in an online fashion. These properties have posed new challenges in terms of efficient storage, analytics, and knowledge extraction out of such data. One of these challenges is online cluster analysis, where the goal is to unveil hidden patterns of collective behaviour from streaming trajectories, such as co-movement patterns [4, 9, 10, 13, 35]. What is even more challenging is predictive analytics over mobility data, where the goal is to predict the future behaviour of moving objects, which can have a wide range of applications, such as predicting collisions, future encounters, traffic jams, etc. At an individual level, a typical and well-studied example of such analytics is future location prediction [11, 24, 25, 29, 33], where the goal is to predict the future location of a moving object, given a look-ahead time. However, prediction of future mobility behaviour at a collective level has not been addressed at the degree of its individual counterpart. Concerning the definition of co-movement patterns, there are several approaches in the literature, such as [4, 9, 10, 13]. However, most of the above are either offline and/or operate at predefined temporal snapshots that imply temporal alignment and uniform sampling, which are not realistic assumptions. For this reason, we adopt the approach presented in [35], which, to the best of our knowledge, is the first online method for the discovery of comovement patterns in mobility data that does not assume temporal alignment and uniform sampling. The goal in [35] is to discover co-movement patterns, namely Evolving Clusters, in an online fashion, by employing a graph-based representation. By doing so, the problem of co-movement pattern detection is transformed into identifying Maximal Cliques (MCs) (for spherical, Flock-like clusters) or Maximal Connected Subgraphs (MCSs) (for densityconnected, Convoy-like clusters). In particular, the problem that we address in this paper is the Online Prediction of Comovement Patterns. Informally, given a look-ahead time interval Δt, the goal is to predict the clusters of moving objects that are anticipated to be shaped after Δt. Figure 1 illustrates such an example, where in blue and green we have the information at hand (past and current locations / clusters, respectively) whereas in orange we have the predictions. The problem we address is quite challenging, since, apart from the inherent difficulty of predicting the future, we also need to define how the error between the actual and the predicted Fig. 1  Predicting evolving clusters via trajectory prediction – blue, green, and orange graphs correspond to past, current, and future (i.e., predicted) evolving clusters, respectively 13 GeoInformatica clusters will be measured. This further implies that a predicted cluster should be correctly “matched” with the corresponding actual cluster, which is not a straightforward procedure at all. To the best of our knowledge, the problem we aim to address has not been addressed in the literature yet. Several mobility-related applications could benefit from such an analytics task. In the urban traffic domain, predicting co-movement patterns could assist in detecting future traffic jams, which, in turn, could help the authorities take the appropriate measures (e.g. adjusting traffic lights) in order to minimize them. In the maritime domain, a typical problem is illegal transshipment, where groups of vessels move together “close” enough for some time duration and with low speed. It becomes obvious that predicting co-movement patterns could help in predicting illegal transshipment events. Our main contributions are the following: • We provide an efficient solution to the problem of Online Prediction of Co-movement Patterns. • We propose a co-movement pattern similarity measure, which helps us “match” the predicted with the actual clusters. • We exploit on MONIC/FINGERPRINT [22, 27], a state-of-the-art cluster evolution discovery method, in order to track the clusters’ evolution through time. • We extensively evaluate the aforementioned framework using large-volume real-world data from two different mobility domains, namely, maritime and urban transportation. The rest of the paper is organized as follows. Section 2 discusses related work. In Section 3, we formally define the problem at hand. Subsequently, in Section 4 we propose our methodology and we introduce a co-movement pattern similarity measure along with the cluster “matching” approach. Section 5, presents our experimental study and, finally, in Section 6 we conclude the paper and discuss future extensions. 2 Related work The work performed in this paper is closely related to three topics, (a) co-movement pattern discovery, (b) future location prediction, and (c) co-movement pattern predic (...truncated)