Article Preview
TopIntroduction
Discovering interesting or useful patterns from the spatiotemporal database is called spatiotemporal data mining. In this knowledge discovery process, special importance is given to the underlying space and time entity associated with the data. With the massive spatiotemporal data generated by the Internet Of Things (IoT), remote sensing devices, smart devices, and so on, the need for automated knowledge discovery from these data has increased.
Spatiotemporal co-occurrence pattern (STCOP) mining is the process of discovering Spatiotemporal (ST) events that are present in close geographic and temporal proximity (Shekhar et al., 2015). These techniques can be applied to crime detection, traffic control, location-based services, and so on. For instance, in the traffic control domain, STCOP can be used to find an association between road accidents and climatic changes. Prior knowledge about the sequence of accidents due to a climatic condition could help traffic officials plan prevention activities. Similarly, crime activity sequences like a bar closing followed by an assault and drunk driving could be discovered in the crime domain. In some cases, patterns will be frequent only in certain areas. The patterns mentioned above will be recurring only in areas where bars and highways co-exist. So, regular pattern mining algorithms will not discover these patterns as their support will be low. Identifying such crime sequences or patterns will help officials in taking necessary preventive actions. Most existing works find patterns in the entire database, but location-centric or locally frequent patterns could be more helpful. In this paper, a novel approach for finding locally frequent crime patterns is proposed.
Based on the temporal ordering of the events, STCOP techniques can be classified into:
- •
Unordered patterns: The temporal ordering is not taken into account.
- •
Partially ordered patterns: The events occur in a series of stage.
- •
Totally ordered patterns: The events occur in a sequence, with one event leading to another. They are otherwise called Spatiotemporal event sequence (STES).
A Spatiotemporal sequence is of the form , representing a chain reaction from event type to event type and then to event type until it reaches event type Where represents the total number of events. STES is the process of finding frequently occurring sequences using the follow relationship among all the event instances. The spread of an epidemic can be traced by the sequence of events that caused it. ST sequence mining can be applied in various domains like solar physics, biomedical science, weblog sequence, DNA sequence, gene structure, targeted advertising, location prediction for taxi services, urban planning, etc.
Though spatiotemporal pattern mining could be applied to various domains, this paper focuses on finding geo-spatiotemporal patterns for crime data set in a specific geographic region or sub-region. The main challenge of discovering spatiotemporal sequences in a spatiotemporal dataset is that the events are not uniformly distributed. Hence STES mining on a dataset could not give specific sequences that are significant in a limited region on a limited time interval. The approach proposed in this work can find locally important patterns, overcoming the drawbacks in the existing system.