An Adaptive Machine Learning Framework for Predicting and Enhancing Student Performance in STEM Education

The advent of machine learning in the educational environment opens a number of opportunities for the optimization of the teaching process and the improvement of the results of training in STEM subjects. In this work, we propose a comprehensive application-oriented machine learning framework to predict student academic achievement and apply personalized interventions in STEM education. With the help of a well-crafted dataset consisting of records of academic and behavioral performance of 2,500 undergraduate students for three consecutive academic years, we design and validate a hybrid ensemble approach which uses a combination of Random Forest and Gradient Boosting algorithms. The framework reaches the level of prediction accuracy of 94.2% in detecting at-risk students and an AUC-ROC value of 0.972. Our method outperforms all individual base models. Apart from the predictive task, our framework offers the use of an adaptive recommendation engine suggesting personalized educational resources to students. The proposed framework operates in the form of web-based analytics dashboard for educators. This work highlights how machine learning approaches can be successfully utilized in educational practice, thus bridging the gap between research in education and practical applications of educational technologies.