An Effective Feature Selection Method Using Dynamic Information Criterion

With rapid development of information technology, dimensionality of data in many applications is getting higher and higher. However, many features in the high-dimensional data are redundant. Their presence may pose a great number of challenges to traditional learning algorithms. Thus, it is necessary to develop an effective technique to remove irrelevant features from data. Currently, many endeavors have been attempted in this field. In this paper, we propose a new feature selection method by using conditional mutual information estimated dynamically. Its advantage is that it can exactly represent the correlation between features along with the selection procedure. Our performance evaluations on eight benchmark datasets show that our proposed method achieves comparable performance to other well-established feature selection algorithms in most cases.