PSO with crossover operator applied to feature selection problem in classification
Houassi Hichem, Rafik Mahdaoui, Toufik Messaoud Maarouk · 2018
In recent years, there is a large number of features in datasets used in pattern classification, which include relevant, irrelevant, and redundant features. However, irrelevant and redundant features decrease the computational time and reduce the classification performance. Feature selection is a preprocessing technique that which to choose a sub-set of relevant features to achieve a similar or even better classification performance than using all features. This paper presents two new hybrid algorithms for a feature selection called particle swarm optimization with crossover operator (denoted as PSOCO1 and PSOCO2); the algorithms are based on the integration of a particle swarm optimization (PSO) and a crossover operator (CO) of the genetic algorithms. A new relevant features vector (RFV) is introduced and used by our algorithms for execute a crossover operator between the RFV and other features vectors. To demonstrate the effectiveness of these algorithms, we compared them with standard PSO [14], PSO4-2 [8] and HGAPSO [28] on twelve benchmark datasets. The results show that the two proposed algorithms significantly reduce the number of selected features and achieve similar or even better classification accuracy in almost all cases.