A Context-Based Association Rules Mining Method for Multiple Event Sequences

Association rules mining of event sequences aims to discover interesting patterns of different neighboring events and plays an important role in understanding their mutual relationship. However, for most existing methods, the distribution characters of events in the sequences are usually ignored and selecting proper thresholds is really a tough task, which brings about the problems of redundant results or interesting rules missing. Thus, new measuring indexes were defined and a context-based method for multiple event sequences mining was proposed. Results of both the simulated experiment and practical cases emphasized that the proposed method could effectively reduce the redundancy in the results in comparison with the classic MOWCATL method. Moreover, there was good consistency between the measuring indexes, which eases the selection of generated rules. Finally, the proposed method was applied to mine association rules between and PM_2.5 concentration and several meteorological factors. Results indicated that the most associated meteorological factor with PM_2.5 concentration was the humidity and an eligible environment for high PM_2.5 concentration were high humidity, low temperature and weak winds.