A Genetic Algorithm for Rule Generation to Predict Student Success

This research used a Genetic Algorithm to generate rules to predict student success. This is a challenging problem for many reasons including limited datasets, unknown attributes needed to make predictions and others. A dataset of the University of California Data Repository was used. The MinMax Scaler algorithm was used for pre-processing and a Genetic Algorithm individual gene mapping was created. A rule with a condition, greater than or less than, and a value is part of the gene mapping. The fitness function was defined as the sum of the percent of the times that each rule was correct. The Genetic Algorithm was run with 100 individuals for 100 generations. The fitness rose from around 20 to level off slightly above 22, with an optimal value located with a fitness of 24.2.Rule generation is important for future research on this problem. Attributes that are no relevant to success predictions are identified. Rules can be used to construct algorithms for predictions. Single rules by themselves or rules in combination can be used to make predictions. This research extends the knowledge of predicting student success.