Genetic-based K-means algorithm for selection of feature variables

This paper proposes a genetic-based K-means (GK) algorithm for selection of the k value and selection of feature variables by minimizing an associated objective function. The algorithm combines the advantage of genetic algorithm (GA) and K-means to search the subspace thoroughly. Therefore, our algorithm converges globally. A weighting junction is then introduced to initialize the parameters of the algorithm. The experiments on a synthetic dataset and a real dataset shows that (i) GK outperforms K-means since GK achieves the minimal value of the objective junction and (ii) GK with the weighting function performs better than GK