DROPOUT PREDICTION USING IMPROVED CONVOLUTIONAL NEURAL NETWORKS

When students who are enrolled at an institution, discontinue their studies before earning a diploma or certificate, this is called a student dropout. It is a major note of concern in schools across the globe because of the negative impact it hason people and communities. This work aims to build an integrated system for dropout prediction in order to solve the significant problem of student dropout rates. The study's overarching goal is to improve dropout prediction accuracy by making use of state-of-the-art methods in preprocessing, feature selection, and classification. We have used an updated standard scalar for preprocessing to manage outliers and promote overall data standardization. The dataset used in this research was sourced from the Kaggle repository. To successfully decrease dimensionality and find the most significant features, a mix of PCA and RFE was used for feature selection. An Improved Convolutional Neural Network (ICNN) algorithm, which incorporates architectural and training methodology enhancements to increase prediction accuracy, was used for the classification phase. Applying the suggested technique to a real-world dataset has revealed its efficacy in forecasting student dropout, as it displayed better performance. Specifically, the Improved CNN algorithm achieved a remarkable 99.67% accuracy rate, which is far higher than that of conventional models, thereby demonstrating its effectiveness in the prediction of children who are most likely to drop out of school. Presenting a thorough framework for dropout analysis that incorporates state-of-the-art approaches in data preparation, feature selection, and classification, this paper makes a significant contribution to the area. A new strategy for forecasting student dropout using an Improved CNN algorithm has emerged, demonstrating improvements in accuracy over more traditional approaches.