Mutual information based feature selection for symbolic interval data

Feature selection is an important issue for data analysis. But almost all feature selection algorithms in machine learning have the weakness to process symbolic interval data because it is structured. In this paper we propose a feature selection method for interval data based on mutual information. The concept of heuristic mutual information is defined as a measure of relevance among features. Experiments on eight data sets show that the proposed method outperforms the previous ones both in accuracy and other criteria.