PTCR-Miner: An effective rule-based classifier on multivariate temporal data classification

Multivariate temporal data are hybrid data. Numeric and categorical data type could be consisted of. Most past researches cannot be operated directly on the multivariate temporal data with both types. Additionally, no useful and readable rules are provided in their methods for advanced classification analysis. We proposed Progressive Temporal Class Rule Miner (PTCR-Miner) algorithm to achieve the classification on multivariate temporal data with a rule-based designed. Through our algorithm, all really useful classification rules are discovered. The rules follow the purification concept we defined, which makes rules comprehensible and intuitive for general users on data classification. We did several experiments to evaluate our method with a multivariate temporal data simulator. Experimental results show PTCR-Miner performs effectively and efficiently on the different simulated multivariate temporal datasets. That means the discovered rules are really helpful and comprehensible for data classification. Further-more, the rule-based and flexible architecture enables PTCR-Miner more applicable to different areas of multivariate temporal data classification.