A Parallelized Frequent Temporal Pattern Mining Algorithm on a Time Series Database

For many years, time series data have been considered as one of the most popular significant data forms in our daily life. Time series exist in many application domains such as finance, medicine, geology, meteorology, and telecommunication. Among the time series mining tasks, frequent temporal pattern discovery is an interesting task because this task brings us a deep insight view of relationships between many objects and events through time. However, it is challenging when a combinatorial explosion occurs with many longer time series. It is more challenging if more informative patterns are required from a time series database. In this paper, we propose a parallel algorithm, called PTP, to cope with the frequent temporal pattern mining task. Our PTP is developed with multithreading on a frequent temporal pattern tree where each branch is processed in parallel. In addition, our PTP maintains the details of each frequent temporal pattern not only from its frequent occurrences in an individual time series but also from its frequent inter-time series associations. As a result, frequent temporal patterns discovered by PTP are more informative with explicit and exact temporal information, showing the relationships among the objects/events corresponding to the time series. Through the experimental results on real time series, PTP outperforms the brute force algorithm and the existing non-parallel algorithm in terms of both time and space. The found frequent temporal patterns can be further analyzed for other tasks such as prediction, classification, and clustering.