Enhanced feature selection models using gradient-based and point injection techniques

This paper focuses on enhancing the effectiveness of filter feature selection models from two aspects. First, feature-searching engine is modified based on optimization theory. Second, a point injection strategy is designed to improve the regularization capability of feature selection. The second topic is important, because overfitting is usually experienced. To evaluate the proposed strategies, we implement these strategies to modify two classic filter feature selection models. One model is based on sequential forward search scheme and the other employs genetic algorithms (GA) for feature selection. Comparing the original and modified models on synthetic and real data, the contributions of our modification are shown.