This research seeks to improve the analytical accuracy of sports and health data for university students by utilizing convolutional neural network (CNN) algorithms. It explores the relationship between physical activity and overall well-being, while introducing an innovative dimensionality expansion technique. This approach employs the least squares principle in conjunction with the Kronecker product, transitioning the algorithm toward a more data-driven framework. By integrating a combination of time-frequency and time-distance data as inputs for a CNN-LSTM network, the study facilitates the automatic identification of movement patterns among students from spectral data. The analysis of the data substantiates the effectiveness of the proposed CNN-based system in evaluating the sports and health metrics of college students.