Hybrid Deep Learning for Academic Achievement Prediction Using Spatio-Temporal and Behavioral Data in Higher Education

Accurate prediction of student academic performance is essential for enabling timely and effective educational interventions. Many existing prediction approaches focus either on academic outcomes or behavioral trends, without fully capturing the interaction between spatial performance indicators and their temporal evolution. To address this limitation, this study proposes a hybrid deep learning model that integrates spatio-temporal information for forecasting student achievement in higher education. The proposed framework combines a Convolutional Neural Network (CNN) to extract spatial features from normalized academic performance data with a Long Short-Term Memory (LSTM) network to model temporal patterns in student behavioral attributes, such as attendance and participation. In addition, FOX optimization is applied to adaptively tune the learning rate, improving training stability and predictive performance. The model is evaluated using student academic and behavioral datasets, and its performance is compared with commonly used baseline models. Experimental results show that the proposed CNN–LSTM approach achieves an accuracy of 97.18 per cent, outperforming standalone LSTM and Support Vector Machine (SVM) models. Furthermore, the model effectively classifies students into low, medium, and high academic risk categories, supporting early identification of at-risk students and facilitating timely intervention in higher education environments.