Assessing and predicting student performance is crucial forboth universities and students, enabling early interventions to mitigate academic risks. Traditional machine learning approaches typically represent student records as one-dimensional feature vectors, which struggle to capture the chronological structure and interdependencies of courses in credit-based programs. To address this, we propose a novel image-based representation of student records, where columns represent courses and rows correspond to semesters, preserving the temporal sequence of academic progress. Additionally, we employ term frequency–inverse document frequency (TF-IDF) techniques to identify and position highly correlated courses adjacent to each other. A convolutional neural network (CNN) is then trained on these images to predict academic outcomes. Evaluated on a dataset of more than 3,000 students recorded over 10 academic years, our model achieves 92% accuracy in identifying students at academic risk, significantly outperforming traditional machine learning models, which reach only 77% accuracy.