Efficient Feature Selection Algorithm Based on Population Random Search with Adaptive Memory Strategies

The effectiveness of classifier training methods depends significantly on the number of features that describe a dataset to be classified. This research proposes a new approach to feature selection that combines random and heuristic search strategies. A solution is represented as a binary vector whose size is determined by the number of features in a dataset. New solutions are generated randomly using normal and uniform distributions. The heuristic underlying the proposed approach is formulated as follows: the chance for a feature to be included into the next generation is proportional to the frequency of its occurrence in the previous best solutions. For feature selection, we have used the algorithm with a fuzzy classifier. The method is tested on several datasets from the KEEL repository. Comparison with analogs is presented. To compare feature selection algorithms, we found the values their efficiency criterion. This criterion reflects the accuracy of the classification and the speed of finding the appropriate features.