Deep Feature Screening Method Based on a Cascade Algorithm

Abstract

To improve the recognition speed and reduce the affection of the redundancy information in pattern recognition, the features should be screened to move those features that have smaller influences. A new joint feature screening method is proposed. A clustering-based dispersion ratio algorithm is used to screen features initially in order to remove those features which have worse intra-class consistency and interclass difference. Then an improved genetic algorithm is employed to deeply screen features and obtain the candidate feature subsets. At last, the inferred statistics is applied to obtain the support of each feature and the best feature subset can be obtained according to the supports. The experimental results show that the joint screening method proposed in this paper can improve not only the classification speed but also the recognition rate.