Investigating the Impact of Similarity Metrics in an Unsupervised-Based Feature Selection Method
Carine A. Dantas, Romulo de O. Nunes, Anne M. P. Canuto, Jolao C. Xavier-Junior · 2017
This paper presents a study about the impact of evaluation criteria and similarity measures in an unsupervisedbased feature selection (FS) method. The main aim of this paper is to assess the importance of these parameter in the analyzed FS method. This method will be evaluated using eight different configurations of these two important parameters. Basically, different correlation measures will be chosen as evaluation criteria and distance measures as similarity measures. In addition, in order to evaluate the impact of these parameters, an empirical analysis will be performed, in which the different configurations will be evaluated to define the best configuration. Then, the result of the best configuration will be compared to existing feature selection and extraction methods, applied to different classification problems. The results shown in this paper indicate that the proposed method using the best configuration had better performance results than the existing methods, in most cases.