Predicting Student Dropout Rates Using Exploratory Data Analysis with Graph Attention Network based on Attention Mechanism

In recent days, student dropout has become a major challenge for various schools and colleges. However, high dropout rates lead to significant economic and social consequences, including wasted resources and diminished competitiveness, ultimately affecting individuals, institutions. The Machine learning (ML) and Deep Learning (DL) algorithms gained more popularity in current scenarios and presented many techniques to address this problem. In this research, an Exploratory Data Analysis (EDA) with Graph Attention Network (GAT), which is named as (EDA-GAT) is proposed for effective student dropout rate. Initially, the input data is collected from the HarvardX Person-Course Academic Year 2013 DE Identified (HMPC) dataset and then preprocessed with the help of one-hot encoding and data imputation techniques to remove the empty rows and missing values. After that, the complex patters are found by the EDA to evaluate the dropout causes of students. Next, these patterns are further processed with GAT to predict the dropout rates by using coefficient multi-head attention and feed forward network. From the results, the proposed EDA-GAT established better results in terms of accuracy (99.7%), precision (98.4%), recall (99.6%) and F1-score (99.1%) when compared to existing Deep Factorization Machines (DeepFM) model.