Monitoring of causal relationships on data stream using time segment characteristic

Vast numbers of stream data are obtained in various fields. Generally, experts in each field monitor if specific state transitions appear in the data stream. The paper proposes a method to detect characteristic state transitions from stream data. The method represents a feature of a state transition as a state transition pattern with a sequence of time segments which have respectively specific conditions. The state transition pattern can flexibly represent characteristics of a state transition. Experts can specify a state transition pattern with some past state transitions. In a data stream, a past state transition affects a state transition in the future. This is a causal relationship. The method the paper presents represents a causal relationship among state transition patterns as a rule. The paper also proposes an active stream database system using the method. This system can pursue multiple possibilities which are due to monitoring the data stream. In an experiment using the data of a thermal power plant, 89.96% of all state transitions were detected correctly.