A New Feature Evaluation Algorithm and Its Application to Fault of High-Speed Railway

Multi-criterion feature ranking algorithms can ease the difficulty on selecting appropriate ranking criterion caused by single-ranking algorithms, and improve the reliability of feature ranking results. However, the issue of conflict between different single-ranking algorithms is often overlooked. By treating this task as a search and optimization process, it is possible to use the D-S theory and evidence conflict to reduce conflicts between different single-criterions and improve the stability of feature evaluation. This work presents a new multi-criterion feature ranking algorithm based on D-S theory and evidence conflict theory combining different criteria improving classification performance of feature selection results. Comparison between the new algorithm and Borda Count, Fuzzy Entropy, Fisher’s Ratio and Representation Entropy methods are done on train fault dataset. The obtained results from the experiment demonstrate that the new algorithm has highest classification accuracy than the other four criterions on all cases considered.