An improved differential evolution and its application to determining feature weights in similarity-based clustering

In this work, we propose an optimization model to tune feature weights for improving performance of clustering via a minimization of uncertainty (fuzziness and non-specificity) of its similarity matrix among objects. To solve the proposed model efficiently, we propose an evolutionary search approach by integrating multiple strategies from both differential evolution and dynamic differential evolution. Then, the proposed method is applied to both weighted fuzzy c-means and weighted similarity-matrix-based transitive closure clustering. Experiments on 11 benchmarking databases show that the proposed method outperforms clustering methods without feature weighting and the feature weighting method based on gradient descent in terms of clustering performance evaluation indices and robustness.