Research on Prediction and Optimization Strategy of University Wi-Fi Network Access Quality Based on Machine Learning

: University Wi-Fi networks are characterized by high user density, diverse service types, and uneven spatiotemporal distribution of access behaviors. Traditional threshold-based and experience-driven network operation methods struggle to achieve precise prediction and dynamic optimization of access quality. This study collects multi-dimensional data from a university campus Wi-Fi network, employing a hybrid machine learning prediction method combining XGBoost and LSTM to accurately forecast QoE metrics including Wi-Fi access rate, latency, packet loss rate, and connection success rates. Additionally, a closed-loop optimization scheme of "prediction-scheduling-evaluation" is proposed, encompassing AP power adaptation, automatic channel configuration, load balancing, and service tiered scheduling. Experimental results demonstrate that the proposed prediction model achieves an average MAE below 5%. The optimization strategy enhances average network access rate by 28.3%, reduces latency by 34.7%, and improves connection success rates to over 99.5%, providing a practical solution for intelligent operation and quality improvement of university Wi-Fi networks.