Feature Selection Methods Based On Improved Particle Swarm Optimization

With the continuous development of information technology,more and more high-dimensional data sets have been produced in the field of data mining.These data sets often involve a large number of irrelevant,redundant,and noise features that affect the classification model and reduce the performance of the learning algorithm.Feature selection as a data dimension reduction technology can effectively solve this problem.Feature selection improves classification performance by eliminating irrelevant and redundant features,and avoids the pitfall of dimensionality curse.There are many existing feature selection methods,including particle swarm optimization algorithm.Because of its simplicity,few tunable parameters and high efficiency,using particle swarm optimization algorithm for feature selection has been a research hotspot now.In this study,the feature selection method based on Particle Swarm Optimization(PSO)is well studied and two improvements are made.Firstly,the algorithm itself is improved.Second,the fitness function is redesigned.First,we analyze the defects of the binary quantum particle swarm optimization algorithm in the process of feature selection,and propose a feature selection method based on improved binary quantum particle swarm optimization algorithm.The new algorithm uses a complete learning strategy to calculate the local attractors.In order to prevent the particles from falling into local extremum in the later iterations,the crossover and mutation ideas of the genetic algorithm are introduced to enrich the diversity of the population before the end of each iteration.For the design of the fitness function,four criteria are used: classification accuracy,the number of features,correlation between features and class,and correlation between features.The correlation is calculated from the maximal information coefficient.The final experiment proves that our proposed algorithm can obtain better classification results by selecting fewer features during feature selection.Furthermore,we analyze the problems existing in the feature selection using the binary particle swarm optimization algorithm and propose an improved adaptive binary particle swarm optimization algorithm.For the update of the individual optimal position and the global optimal position,the two indicators of fitness value and the number of features are considered.Some improvements have also been made to the update mechanism of particle positions.And a mutation mechanism was introduced to enrich the population before the end of each iteration.The design of fitness function not only considers the classification accuracy and the number of features,but also introduces the contribution of the features to the classification.The degree of contribution is calculated from mutual information.The final experiment shows that our proposed algorithm shows better performance in feature selection.