Research on the Influence Factors of Teacher Role Training Based on Recurrent Neural Network

Based on the theoretical results of the research on educational supervision, this paper establishes a framework suitable for the research on the training of university teachers so that practical theories guide the supervision practice of grass-roots university teachers and carries out theoretical research in the aspects of the supervision organization, the composition of responsible university superintendents, and the connotation of supervision, which has specific complementary value and exploration significance for the research on the training of university teachers. This paper proposes a method of factor feature selection based on recurrent neural networks to solve the problem of some features of “teacher role” data not being very relevant to the prediction task. The method first uses a random forest feature selection method to train the original data. Then, each factor is perturbed, the data after the disturbance is obtained, the same benchmark model is used for training, and the fact score is calculated as an index to measure the importance of the feature. This method based on recurrent neural networks effectively solves the problem of feature selection, ignoring the instability of feature interaction and importance score in random forests.