Adaptive Feature Selection for Enhanced Predictive Modelling in Educational Data Mining

Through Educational Data Mining (EDM) users can derive academic decisions from educational data through pattern recognition processes. Educational datasets become difficult for predictive models to work effectively due to their combination of complexity with high dimensionality. AFSA represents an Adaptive Feature Selection Algorithm that works to boost model performance through feature selection of important variables and reduction of unneeded features and unpredictable data points. AFSA implements Information Gain and Chi-Square and Gini Importance for feature selection quality during real-time operation. A Genetic Algorithm applies at runtime to enhance the feature subset selection by optimizing performance metrics which combine accuracy and computational speed. Laboratory tests performed on multiple educational datasets prove that the framework diminishes the number of variables without compromising achievement levels. The successful practical implementation of this framework as an educational analytics tool can be justified by achieved results. The presented adaptive-hybrid feature selection method creates effective data prediction models through efficient methods to enhance educational outcomes and decision-making.