A Hybrid Feature Selection for MRI Brain Tumor Classification

Because a great number of features affects the performance of classification systems, a growing emphasis is placed on the feature selection. This work seeks to obtain an optimal feature subset through a hybrid algorithm of Simulated Annealing-Genetic Algorithms (SA-GA). Our proposed approach mutually avoids being stuck in a local simulated annealing minimum with the very high convergence rate of the genetic algorithm crossover operator and thus guarantee a high computational efficiency of support vector machine. To evaluate the proposed approach, a real dataset of brain tumor Magnetic Resonance Images was used. The proposed approach was compared to the methods of simulated annealing and a genetic algorithm used separately. The obtained results showed that SA-GA outperforms simulated annealing and genetic algorithms when they are applied in isolation, in terms of accuracy and computing time.