Machine Learning-Based Prediction in HTTP Request–Response Cycles: Impacts on Webpage Quality Metrics

Abstract

The hypertext transfer protocol (HTTP) request–response cycles during webpage access and content posting exhibit recognisable patterns; however, no unified standard currently streamlines both activities, despite the existence of independent specifications for each. Previous research has leveraged cycles of client–server requests and responses to predict outcomes such as user behaviour (UB) analysis, anomaly detection (AD), performance optimisation (PE), predictive maintenance (PM) and user authentication and security (UA), often without explicitly associating these activities. Addressing this gap, the present study focuses on the combined modelling of HTTP request–response cycles for both webpage access and personal information submission. An experimental study was conducted, where HTTP sessions were generated and analysed for both access and posting activities. Six machine learning models—Decision Tree, Random Forest, Gradient Boosting, k-Nearest Neighbours (kNNs), Logistic Regression and Support Vector Machine—were applied to both the CSIC 2010 HTTP dataset and lab-generated HTTP transmission datasets across the UB, AD, PE-PM and UA tasks. Results indicate that the Random Forest classifier achieved the highest accuracy of 97.53% in predicting AD-based HTTP request–response cycles during webpage access, and 85.93% accuracy in predicting PE-PM tasks during content posting. Gradient Boosting, kNNs and Support Vector Machine models also demonstrated strong versatility and robustness across different HTTP cycle prediction tasks. Furthermore, the analysis concluded that HTTP request–response cycles for webpage access exhibit greater structural consistency compared to those associated with content posting activities.