Hybrid algorithms of microbial genetic algorithm and particle swarm optimization for automatic learning groups composition

Abstract

Collaboration learning within diverse groups of varying knowledge levels and varying interest levels have been noted to improve learning outcomes. However, composing balanced learning groups with diversity in knowledge level and interest level within the groups while maintaining similarity among the groups is NP-hard and time consuming. The primary aim of this research is to develop an algorithm for automatic composition of balanced learning groups in (MOOCs) with minimal human intervention. The algorithm will assist facilitators in forming balanced learning groups with ease for learners in online classes to benefit from effective collaboration. The research design was experimental research. This design help established comparative experiments of the new algorithm with the particle swarm algorithm as the bench mark algorithm. The findings in the first experiment showed that, the hybrid MGAPSO (Microbial Genetic Algorithm and Particle Swarm Optimization) algorithm outperformed the PSO (Particle Swarm Optimization), an ANOVA (one-way test) showed high significant difference in the mean fitness of the two algorithms (hybrid Microbial Genetic Algorithm and Particle Swarm Optimization and Particle Swarm Optimization). A possible explanation for this might be that the microbial genetic algorithm component tends to re-introduce new particles at every iteration after every genetic operation, thus, introducing diversity in the swarm. In the second experiment, the new adaptive hybrid AMGAPSO (Adaptive hybrid Microbial Genetic Algorithm and Particle Swarm Optimization ) outperformed both the PSO (Particle Swarm Optimization) and the hybrid MGAPSO (Microbial Genetic Algorithm and Particle Swarm Optimization) with high significant difference in the mean fitness of the new adaptive hybrid AMGAPSO (Adaptive hybrid Microbial Genetic Algorithm and Particle Swarm Optimization ) and the mean fitness of the hybrid MGAPSO (Microbial Genetic Algorithm and Particle Swarm Optimization) and that of the Particle Swarm Optimization. A possible explanation of this finding is that particles stuck in the location in the PSO with their re-initialised new velocity may have searched the solution space in different directions and may have jumped out from their respective locations using the microbial genetic algorithm component, which suggest that the method of hybridisation could have resulted in the improved performance of the adaptive hybrid AMGAPSO algorithm relative to the hybrid MGAPSO. In the third experiment 500 learners profile data was used in a comparative experiment of the adaptive AMGAPSO (Adaptive Microbial Genetic Algorithm and Particle Swarm Optimization), MGAPSO (Microbial Genetic Algorithm and Particle Swarm Optimization) and PSO (Particle Swarm Optimization), groups formed by the new adaptive AMGAPSO algorithm were analysed; The ANOVA (one way) test results showed no significant difference in the means of all groups for all six learners attributes among all groups formed by the algorithm. The understanding at this point is that, the adaptive hybridization method may have provided a means of avoiding the problem of parameter adjustment and the fitness function derived have contributed to the formation of groups with diversity within the groups while maintaining similarity among the groups formed by the algorithm. The findings answered the aim of the research as the new algorithm outperformed the existing algorithm used in the literature and the groups formed with the new algorithm were balanced in all profile features used in all the groups. The findings show that the new algorithm can be used to form balanced learning group with minimal human intervention. Limitations of the research are that only 500 learners’ data was used for the validation of the experiment; this was because data for more students who belong to the same class could not be obtained; ANOVA (One Way Test) was the only statistical tool used in the analysis. In addition, the algorithm developed was evaluated only in terms of the groups form however the effectiveness of the groups and the overall learning improvement achieved by the groups formed by the algorithm was not evaluated.