Short videos are increasingly being consumed by college students as crucial content in the age of big data since they are a perfect fit for this medium. Therefore, college students should place a high value on the utilization of short movies. In this study, a neural network is utilized to create a mental health prediction model for college students. The neural network is trained using its self-learning capability to map out the relationships between different elements and mental health. The enhanced algorithm minimizes the production of candidate item sets to some amount, as well as the algorithm's time and space requirements, significantly decreasing the initialization time of the transaction set. According to the research, the test sample's pattern recognition accuracy was 81.29%, whereas the training sample's accuracy for pattern recognition was 83.37%. The analysis's finding is that the enhanced mining algorithm offers a fresh approach to educating college students about their health.