Abstract: The poverty alleviation work for college students has always been the focus of attention in education. How to use effective big data analysis methods to reduce the workload of review and fair review process and achieve the goal of targeted poverty alleviation in colleges and universities is a question worthy of further study. Based on the behavioral data of college students, this paper combines the time-series characteristics of college data, extracts the basic information and multi-dimensional features of behavioral data, and proposes a clockwork long short-term memory (CW-LSTM) algorithm based on deep learning theory for prediction. Finally, the model is verified using real data, and the results show that our method is better than the Naive Bayes algorithm and decision tree algorithm.