Using Integrated Machine Learning Models On Educational Big Data To Predict Student Dropout Risk

Abstract

Student dropout remains a significant issue in education, impacting career opportunities and economic burdens. Despite predictive analytics' growing interest, current methods struggle to handle complex datasets effectively. This research addresses that gap by proposing an integrated Machine Learning (ML) framework for predicting student dropout risk using educational big data. The model utilizes a comprehensive dataset collected from institutional academic records, attendance logs, demographic profiles, and behavioral metrics curated to ensure representation of various influencing factors. Data preprocessing involves missing values (MVs) handling, normalization, and outlier removal using robust statistical techniques. For dimensionality reduction and effective feature extraction, Kernel Principal Component Analysis (Kernel PCA) is employed, capturing non-linear relationships within the data to enhance model interpretability and performance. The core novelty of this framework is the proposed Efficient Locust Swarm-tuned Dynamic Extreme Gradient Boosting Classifier (ELS-DXGBoostC), which dynamically adjusts learning rates and tree parameters, optimized through nature-inspired Locust Swarm optimization to enhance predictive accuracy and generalization. Experimental results show superior performance in terms of accuracy (98%), recall (98.4%), precision (97.5%), and F1-score (97%) compared to traditional classifiers. This work integrates swarm intelligence with gradient boosting, enhancing the academic field and supporting institutions in deploying scalable, data-driven solutions to reduce student attrition.