Finding spatiotemporal co-occurrence patterns of heterogeneous events for prediction

Abstract

Advances of IoT facilitates leverage of heterogeneous sensing data over the Internet, such as remote sensing data, traffic data and SNS data. Integrated analysis of IoT data is crucial part for urban emergency management in smart cities in order to predict various social events co-occurring with a natural disaster event. Discovering of spatiotemporal co-occurrence patterns is a task of integrated analysis of IoT data and has received a lot of attention. However, such spatiotemporal co-occurrence patterns can fail to capture local events that occur in limited regions and limited time intervals. In this paper, we consider the problem of mining spatiotemporal co-occurrence patterns from IoT sensing data, each of which is annotated with a valid spatial and temporal region. Our idea is to incorporate spatiotemporal clustering with the frequent itemset (pattern) discovery process to reduce spatiotemporal bias of event distributions and we repeat this process in greedy approach in order to capture patterns with difference scales. By this way, our algorithm improves accuracy of the frequent itemsets. We applied our method to discovery and prediction of traffic disaster events co-occurring with torrential rain events in Kansai area, Japan. Our experimental result shows 31% improvement of prediction performance on F-measure against a baseline.