Enhancing Student Dropout Prediction in Educational Data Mining Using Sparse Feedback Based Deep Residual Network With Xception and Optimised Feature Selection

ABSTRACT Student dropout is one of the trickiest and most detrimental problems in education; it has an impact on both students and institutions. Predicting student dropout rates early helps mitigate the negative social and economic effects. In order to solve the issue, this article suggests a novel method for predicting the dropout rate of students. Data transformation, feature selection and student dropout prediction are three steps involved here. Input data is passed to data transformation, which is the process of converting the format or structure of a dataset to match that of a target system. Next, by integrating Serial EWMA concept (SE) in Puffer fish Optimization Algorithm (POA), Serial exponential Puffer fish Optimization Algorithm (SE‐POA) is proposed to determine the significant components. Ultimately, the chosen features are exposed to the student dropout prediction phase. At this point, FbResNet‐Xception, which is the combination of FbResNet and Xception, is used for predicting student dropout. In comparison to conventional models, experimental results show that FbResNet‐Xception performed better, showing an accuracy of 95.9%, dropout recall of 91.1%, dropout precision of 97.3% and dropout F‐measure of 94.1%.