An Improved Early Student'S Performance Prediction Using Deep Learning

Presently owed to technological revolution, a massive amount of data is generated in every field. Extracting the useful insights from consequential data is a very critical task. Moreover, advancement in the Deep Learning (DL) models ensued in the effective prediction of data. In the proposed study DL model is used for predicting the student's academic performance. Early prediction of the student's performance will reduce the risk of failure. The study used two courses data, i.e., mathematics and Portuguese language, containing demographic, socio-economical, educational and student's course grades data. The data set suffers from the imbalance; SMOTE (synthetic minority oversampling technique) is used to overcome imbalance issue. The performance model is evaluated using several feature sets (all features, excluding G2 and G3) and evaluation measures such as precision, recall, F-score, and accuracy. The results showed the significance of the proposed DL model in early prediction of the students' academic performance. The model achieved an accuracy of 0.964 for Portuguese course data set and 0.932 for mathematics course data set. Similarly, the precision of 0.99 for Portuguese and 0.94 for mathematics is achieved. Furthermore, all the features are significant in predicting the student's academic performance.