An Improved Clustering Algorithm Based on Feature-weight Learning

摘要

We propose an improved data clustering algorithm, which consists of Feature-weight Learning algorithm and Clustering algorithm. In our algorithm, the Feature-weight Learning algorithm assigns weight for each feature by minimizing feature evaluation function FLearning(w) according to the gradient descent technique and integrates the weighted feature with Fuzzy C-means clustering. Some merits of the genetic algorithm and the simulated annealing algorithm for data clustering are also adopted in our algorithm, in which the important feature's effect is emphasized, while the redundant feature's effect is weakened, moreover the local optimal solution is resolved. Experiments reveal that the clustering results of our algorithm are better than that of the traditional improved Fuzzy C-means clustering.