Variable Minimization-based Prediction Model with Enhanced Genetic Algorithm using Flip Multi-sliced Average Crossover (FMSAX) Operator

Abstract

The genetic Algorithm (GA) relies on the effective utilization of the crossover operator, which plays a crucial role in mitigating premature convergence, may occur when genetic operators dominate the population too quickly, leading to a loss of diversity and hindering the generation of optimal solutions. This study introduced a new crossover operator called the flip multi-sliced average crossover (FMSAX) operator with a rank selection function to address this problem. This new operator was more efficient for variable minimization in GA. The study used 2,403 records of student attrition from two private higher education institutions in the Philippines and found that the enhanced crossover outperformed the original crossover operator in GA. The improved GA using the FMSAX operator resulted in a 53.33% improvement in variable minimization, with 25% fewer variables eliminated than the original average crossover (AX) operator. Additionally, when integrated with the C4.5 algorithms, the enhanced GA using FMSAX produced an accuracy of the prediction model of 96.47%, higher than the original GA using AX and C4.5 algorithms with an accuracy of 93.85%. This study's prediction model using the enhanced GA and FMSAX operator helped private educational institutions design preventive measures to mitigate attrition rates in the academe.