DEVELOPMENT OF MARINE PREDATOR ALGORITHM (MPA) WITH MUTATION OPERATION FOR FEATURE SELECTION

ABSTRACT

Several metaheuristics algorithm (MA) such as Particle Swarm Optimization (PSO), Salp Swarm Algorithm (SSA) and Marine Predators Algorithm (MPA) have been considered in optimizing Feature Selection (FS) process. MPA as a recent and efficient population based MA has limitations in FS tasks with regards to its exploration/exploitation phase imbalance. This led to the development of more approach for balancing the exploitation phase so as to maximize the classification accuracy and minimize the number of features. The algorithm is combined with a local search to help improve the exploitation phase. In this experiment, random bit flipping was introduced to the last phase of the exploitation phase so as to intensify and refine solution. The experiment uses two different values for mutation rate which are 0.1 and 0.6 respectively, so as to see which give better performance. The experimental results shows that MPA performed better with a mutation rate of 0.1 with an accuracy of 92.4%, sensitivity of 92.6%, specificity of 91.98% and average selected features of 9.3. The standard deviation obtained for accuracy, sensitivity and specificity are 0.005522, 0.004396 and 0.003814 respectively.

Keywords: MPA, Feature Selection, Exploitation, Mutation, Metaheuristics Algorithm