Knowledge discovery based on importance of features

This paper proposes a system which datamines time series classification knowledge leading by a discovery of feature patterns. In the case of classification, prediction accuracy is an important point, and to build a human understandable model is another essential issue. To satisfy these requests, our system runs in two stages. In the first stage, the system discovers important feature patterns which are useful for identifying data. For this purpose, we propose a feature importance measure which is called FI. The second stage builds a decision tree that determines class membership based on the feature patterns. We explain how these two stages are harmonized in the entire process.