Feature weighting using a clustering approach

In recent decades, the volume and size of data has significantly increased with the growth of technology. Extracting knowledge and useful patterns in high-dimensional data are challenging. In fact, unrelated features and dimensions reduce the efficiency and increase the complexity of machine learning algorithms. However, the methods used for selecting features and weighting features are a common solution for these problems. In this study, a feature weighting approach is presented based on density-based clustering. This method has been implemented in two steps. In the first step, the features were divided into clusters using density-based clustering. In the second step, the features with a higher degree of importance were selected in accordance to the target class of each cluster. In order to evaluate the efficiency, various standard datasets were classified by the feature selection and their degree of importance. The results indicated that the simplicity and suitability of the method in the high-dimensional dataset are the main advantages of the proposed method.