Student Mental Health Risk Assessment Based on Federated Learning and Privacy-Preserving Mechanisms: Deep Neural Network Modeling and Analysis of Distributed Big Data

The risk assessment of students' mental health involves two core issues: one is how to efficiently use the mental health data distributed in different institutions (schools, medical institutions), and the other is how to strictly protect the privacy of students when using these data. In response to these problems, this study proposed a DNN modeling and analysis method based on federated learning and privacy protection mechanism. This study constructed a federated learning framework to integrate the mental health data of students scattered on different data nodes. Each node trained the DNN model on the local data set, and the model performed nonlinear transformation on the input data through the hidden layer. After each node completed the local training, the model parameters were encrypted and sent to the central server. After receiving the parameters of all nodes, the central server decrypted and weighted averaged them to generate global model parameters. In addition to the privacy protection provided by federated learning itself, this study also used differential privacy technology to further protect the privacy of students' personal data by adding an appropriate amount of random noise during the model training process. After introducing the privacy protection mechanism, the maximum illegal access rate of the data dropped from 0.099% to 0.052%. The average evaluation time of this research method was 157.08 seconds, which shortened the evaluation cycle and improved the evaluation accuracy compared with the traditional method of 620.4 seconds. Therefore, the method proposed in this study can help improve the efficiency and accuracy of mental health risk assessment while protecting students' privacy, and promote the intelligent development of mental health assessment.