DeepEDM: Deep Learning Based Education Data Mining for Academic Student Performance Evaluation Model

Abstract

Poor grades are one of the biggest causes of school abandonment. This has an influence on performance because so many students find it difficult to adjust to the institution's learning environment once they get there. Other factors include student participation in extracurricular tasks and politics. Learners' performances frequently tend to be unsatisfactory for these different predictable and unpredictable reasons, which have an impact on development. As a result, it's important to examine undergraduate results to identify the real reasons for students' varying levels of performance. The primary goal of our research work is to identify the numerous variables that affect achievement at the under-graduation level. Therefore, the main motivation behind this effort is to help students identify the factors that lead to their performance so that they can take action to change their results. The learners, course teachers, and others will have the opportunity to improve the environment once the major elements have been recognized and assessed. In order to early predict the student’s academic performance, we have proposed a hybrid deep learning model of Recurrent Neural Network – Long Short-Term Memory classifier. This proposed methodology is compared with various traditional machine learning classification models and deep learning classifiers. Using experimental findings, we have observed the classification accuracy of several traditional machine learning techniques like support vector machine, random forest, J48, artificial neural network and naïve Bayes and also deep learning techniques like a deep neural network, RNN. The proposed method of predicting student performance using RNN-LSTM sigmoid, Tanh and ReLU function is performed and the results are compared with the various machine learning and deep learning algorithm. It is observed from the experimental findings that RNN-LSTM (ReLU) outperforms other classifications perform with the highest accuracy rate of 95.5%. Our proposed system produces high classification accuracy when heterogeneous datasets or real-time complicated huge datasets of students with multi-value features are used