Mining fuzzy rules for time series classification

Time series classification is concerned about discovering classification models in a database of pre-classified time series and using them to classify unseen time series. To better handle the noises and fuzziness in time series data, we propose a new data mining technique to mine fuzzy rules in the data. The fuzzy rules discovered employ fuzzy sets to represent the revealed regularities and exceptions. The resilience of fuzzy sets to noises allows the proposed approach to better handle the noises embedded in the data. Furthermore, it uses the adjusted residual as an objective measure to evaluate the interestingness of association relationships hidden in the data. The adjusted residual analysis allows the differentiation of interesting relationships from uninteresting ones without any user-specified thresholds. To evaluate the performance of the proposed approach, we applied it to several well-known time series datasets. The experimental results showed that our approach is very promising.