Design and Application of Intelligent Management System for College Students’ Behavior Detection

Traditional college student behavior detection systems generally rely on a single data source and rule-based analysis methods, resulting in insufficient detection accuracy, poor real-time performance, and lack of personalized management functions. In addition, traditional systems have significant defects in privacy protection and system scalability. To this end, this study proposes an intelligent management system based on multimodal data collection, intelligent algorithms, and privacy protection mechanisms, aiming to improve the accuracy and real-time performance of behavior detection, and to achieve personalized recommendations and privacy protection, thereby overcoming the limitations of traditional systems. During the research, students’ physiological, behavioral and learning data are collected through sensors, cameras and learning platforms, and real-time data stream processing and multi-source data fusion are performed to ensure the spatiotemporal alignment and noise processing of the data. Subsequently, the data is deeply analyzed using the K-means-based behavior classification model and the ResNet-18 (Residual Networks) + LSTM (Long Short-Term Memory) sentiment analysis module to achieve student behavior recognition, anomaly detection and sentiment analysis. Then, a hybrid recommendation strategy is combined with learning behavior data to provide students with personalized learning recommendations, predict learning progress and automatically generate intervention strategies. Finally, technologies such as k-anonymization, differential privacy and data encryption are used to ensure data privacy protection and system security to ensure the sustainability of the system in practical applications. Experimental results show that the system in this study performs well in anomaly detection and sentiment analysis, with a detection accuracy of 95.2%, a recall rate of 92.8%, and an end-to-end delay of only 122 ms, which is significantly better than the control group. In terms of privacy protection, the system achieves an average re-identification rate of 2.13% through k-anonymization, which is much lower than the highest average of 8.15% in the control group. At the same time, in large-scale data processing, the system’s response time is 362 ms, and the resource utilization rates are 74% of the CPU (Central Processing Unit) and 48% of the GPU (Graphics Processing Unit), showing good scalability.