MBO: A Novel Memory based Optimizer for Continuous and Discrete Optimization Problem

Abstract

This study introduces a novel metaheuristic approach called Memory Based Optimizer (MBO), which emulates the problem-solving process through multiple stages by utilizing the best solution obtained in terms of memory. Rooted in principles of human psychology, MBO reflects the tendency for individuals to solve problems incrementally, using previous learning to take small steps toward an optimal solution within a limited number of attempts. MBO is first evaluated on ten CEC 2019 Benchmark Functions, and its results are compared with ten other metaheuristic algorithms under similar execution conditions. In its binary form, MBO is also applied as a wrapper for feature selection in supervised machine learning using the K-NN classifier on twelve benchmark classification datasets. The findings indicate significant improvements in average accuracy and optimal feature selection compared to other metaheuristic approaches. As per the simulation results, MBO has outperformed other metaheuristics approaches in 5 out of 10 continuous benchmark functions. Further, the MBO has achieved higher average accuracy in 11 out of 12 datasets, along with better execution times in 9 out of 12 datasets when applied as a wrapper for feature selection tasks, the average improvement in accuracy and F1 score is reported as 2.8746% and 3.1643% when compared with other metaheuristic approaches under similar execution environment further validating its robustness and efficiency across multiple optimization tasks.