Abstract: In colleges and universities, the registration of major transfer is very popular, and it is often difficult to allocate. Therefore, making preparations in advance is important in major transfer. If we can predict the enrollment of major transfer students in that year, it will be of great help to the follow-up work of colleges and universities. Popular college students to professional enrollment of Jilin University from 2003 to 2017 is used to establish the number of popular college students forecast model; the DNN(Deep Neural Network) deep learning network structure is introduced in the Google research and development of tensorflow framework to establish the number of popular college students forecast model; finally, the training data for 15 years is used to predict the number of popular college students in 2020 analysis. The method proposed can better solve the problem of the number of candidates for major transfer in popular colleges, and has a certain guiding significance for the follow-up work.

Key words: changing majors, deep neural network (DNN) network, Tensorflow framework