A new method for feature selection based on intelligent water drops

Mohammad Hossein Khosravi, Parsa Bagherzadeh

Abstract: One of the trending research areas of data mining and machine learning is feature selection. Feature selection is used as a technique for improving classification accuracy of a classifier as well as a more convenient way for visualization of data. In this paper, a new method for feature subset selection, based on intelligent water drops algorithm is proposed. Intelligent water drops algorithm is a metaheuristic algorithm which is inspired from movement of water drops in nature. In the proposed method, a new objective function which is suitable for intelligent water drops algorithm is introduced. The objective function is designed such that the selected feature vector would obtain a good classification accuracy as well as providing a good generalization degree. According to the experiments, the use of proposed approach leads to more accurate results as well as significant reduction in number of features.