Incremental Mining of Frequent Sequences in Environmental Sensor Data
Carlos Roberto da Silveira, Danilo Codeco Carvalho, Marilde Terezinha Prado Santos, Marcela Xavier Ribeiro · The Florida AI Research Society · 2015
The mining of sequential patterns in environment sensor data is a challenging task. Most of sequential mining techniques requires periodically complete data. Furthermore, this kind of data can be incomplete, present noises and be sparse in time. Consequently, there is a lack of methods that can mine sequential patterns in sensor data. In this paper, we proposed IncMSTS, an incremental algorithm for mining stretchy time patterns. The proposed algorithm is an incremental version of the MSTS, a previous not scalable algorithm that mines stretchy time patterns in static databases. The experiments show that IncMSTS runs up to 1.47 times faster than MSTS. When compared to GSP, the literature baseline algorithm for mining sequences, IncMSTS can return 2.3 more sequences and the returned sequences can be 5 times larger, indicating that the sparse time analysis promoted by IncMSTS broadens the mining potential of finding patterns.