Differential Evolution and Genetic Algorithm Based Feature Subset Selection for Recognition of River Ice Types

One of the essential motivations for feature selection is to defeat the curse of dimensionality problem. Feature selection optimization is nothing but generating best feature subset with maximum relevance, which improves the result of classification accuracy in pattern recognition. In this research work, Differential Evolution and Genetic Algorithm, the two population based feature selection methods are compared. First, this paper presents Differential Evolution float number optimizer in the combinatorial optimization problem of feature selection. In order to build the solution generated by the Differential Evolution float-optimizer suitable for feature selection, roulette wheel structure is constructed and supplied with the probabilities of features distribution. To generate the most promising feature set during iterations these probabilities are constructed. Second, Genetic Algorithm minimizes the Joint Conditional Entropy between the input and output variables. Practical results indicate Differential Evolution feature selection method with ten features achieves 93% accuracy when compared with Genetic Algorithm method.