Software-based performance evaluation and forecasting of web applications using machine learning models

This research aims to develop and evaluate a software-based methodology for performance assessment and forecasting of web applications using machine learning models. The proposed approach integrates automated data collection, preprocessing, analysis, and real-time visualization within a unified software framework. The experimental evaluation was conducted on a dataset comprising over 10,000 real-world performance records, including response time, CPU and memory usage, network throughput, and JavaScript execution metrics collected via browser DevTools. Linear and polynomial regression models, a decision tree, and a neural network were applied to identify performance patterns and predict key metrics. The results demonstrate that linear regression achieves the highest overall accuracy (MAE = 187.25, R² = 0.9363), while neural networks provide comparable performance under dynamic workload conditions. The developed software framework enables real-time monitoring and reporting through WebSocket-based visualization and supports the identification of performance bottlenecks. The findings confirm that integrating machine learning into automated performance evaluation improves prediction accuracy and accelerates the detection of performance degradation compared to traditional manual monitoring approaches.