Feature Selection:A New Perspective

Feature selection is one of the process data mining tasks. This process finds the optimal feature subset using machine learning and evaluation criteria. Several techniques are used to find the optimum features in artificial and text databases. In this paper, the machine learning based methods are classified as unsupervised learning, semi-supervised learning and supervised learning, which find the various features in the text databases, web databases and gene databases. The evaluation criteria based methods are categorized as Filter, Wrapper and Hybrid approach, which are employed to discover the optimal feature set in artificial datasets. These approaches are very useful in data mining process for improving the prediction performance, reducing the cost and understanding of the features. These issues are addressed by various techniques using measures like dependent and independent criterion. This survey explores the various feature selection processes and their uniqueness for finding the optimal feature subset in term of accuracy, robustness and efficiency.