SVM-RFE and Mutual Information-based feature selection and Shepard Convolutional Neuron Attention Network for Indian student dropout prediction

In the context of education, the dropout rate of students is increasing in each institution, so it is necessary to find out the students who are at risk level of quitting school before completing their studies. This affects the entire educational results of each school and leads to lower the graduation rates. Hence, a Deep Learning (DL) named Shepard Convolutional Neuron Attention Network (ShCNA-Net) is established to predict dropout rates of an Indian student thereby allowing early intervention and support for the particular Indian student. Primarily, the student performance data is obtained from dataset and fed to feature extraction phase, wherein the features, like personal information, curriculum, and economics are extracted. Next, feature selection is achieved using Support Vector Machines-Recursive Feature Elimination (SVM-RFE) and mutual information. Then, data augmentation is achieved using Synthetic Minority Oversampling Technique (SMOTE)-based oversampling method. Later, the student dropout rate is predicted by ShCNA-Net, which is the combination of Shepard Convolutional Neutral Network (ShCNN) and Neuron Attention Stage-by-Stage Network (NASNet). Furthermore, ShCNA-Net achieved a high accuracy of 91.456%, True Negative Rate (TNR) of 91.375%, and True Positive Rate (TPR) of 92.777% with K-value 9 in predicting the dropout rate of the Indian student.