Improved Relief Weight Feature Selection Algorithm Based on Relief and Mutual Information

As the classic feature selection algorithm, Relief algorithm has advantages of simple computation and high efficiency, but itself is limited to only dealing with binary classification problems, comprehensive distinguishing ability subsets composed former K features selected by often redundant, as cannot select ideal subset. When calculating correlation redundancy between characteristics mutual information, speed slow because computational complexity method’s need calculate probability density function corresponding features. Aiming solve above we first improve weight so that it can be used evaluate a set candidate sets. Then use improved joint information evaluation replace basic problem correlation, Finally, compound based on proposed using heuristic sequential forward search strategy. This effectively small strong characteristics, excellent faster calculation speed.