EARLY PREDICTION OF STUDENT’S ACADEMIC PERFORMANCE USING AN INTERPRETABLE AND DEEP LEARNING BASED TIME-SERIES PREDICTION FRAMEWORK

Educational data mining (EDM) is an active research area in present day’s higher education system. One of the prevalent objectives of EDM is to predict academic performance of students sin their early academic days. The proposed framework is based on Convolutional Neural Network (CNN) and Bidirectional Long-Short-Term-Memory (LSTM). This study has been projected towards the estimation of student’s Cumulative Grade Point Average (CGPA) at the time of graduation based on their previous academic records. A pair of custom activation functions is designed to enhance the performance of the proposed framework. An attention mechanism is implemented to focus on the most dominant features of the prediction process. To interpret the major contributing factors of our study, an interpretable module is used. Looking at the progression of the performance curve, it is observed that student’s CGPA can be predicted at their early academic days using the proposed model. Experimental results show that the proposed model with attention mechanism and pair of custom activation functions outperforms the existing models in terms of Mean Square Error (MSE) and Mean Absolute Error (MAE). Based on the evolution of efficient learning pattern, students are recommended pre-emptive counselling towards meeting the attainment gap.