A New Feature Selection Method Based on K-Nearest Neighbor Approach

In many data analysis tasks, one is often confronted with very high dimensional data. Feature selection is an effective method to solve the problem with high dimensional data. The aim of feature selection is to reduce the number of features used in classification or recognition. This reduction is expected to improve the performance of classification and clustering algorithms in terms of speed, accuracy and simplicity. This paper proposes a new unsupervised feature selection algorithm which is based on K-nearest neighbor approach. The proposed algorithm evaluates the whole features according to the each sample in the dataset by the K-nearest neighbor approach. After that, the overall assessment is given based on the assessment of features for each sample. We evaluate the performance of the proposed unsupervised feature selection algorithm using the well-known UCI machine learning datasets, and the results illustrates the proposed algorithm is comparable with the traditional feature selection algorithm.