Fuzzy Clustering-Based Filter

This paper introduces a filter, named FCF (Fuzzy Clustering-based Filter), for removing redundant features, thus making it possible to improve the efficacy and the efficiency of data mining algorithms. FCF is based on the fuzzy partitioning of features into clusters. The number of clusters is automatically estimated from data. After the clustering process, FCF selects a subset of features from the obtained clusters. To do so, we study four different strategies that are based on the information provided by the fuzzy partition matrix. We also show that these strategies can be combined for better performance. Empirical results illustrate the performance of FCF, which in general has obtained competitive results in classification tasks when compared to a related filter that is based on the hard partitioning of features.