Cosine similarity based filter technique for feature selection

Filter-based feature selection techniques are less complex compare to Wrapper-based feature selection techniques in case of High Dimensional datasets. In this paper, we proposed a filter method feature selection, which is Cosine Similarity-based Filter feature selection Technique (CSF) for High-Dimensional Datasets. In this method, absolute cosine similarity with respect to class label is used to ordering the features and from ordered features list a user-defined number of features is selected. Dataset with selected features is tested for classification accuracy using Multi-classifier system (K-Nearest Neighbor (KNN), Classification and Regression Tree (CART), Naive Bayes (NB) and Support Vector Machine (SVM)). This method is applied to four high-dimensional binary class datasets and obtained accuracy shows that method is either better or equivalent compared to other existing methods.