Machine Learning-Based Performance Evaluation and Memory Usage Forecasting for Intelligent Systems

The Performance Measurement Project at Thompson is a targeted effort to evaluate the performance of applications before they are shipped to production. Understanding the importance of performance in designing user satisfaction and experience, the project focused on identifying and resolving performance issues early in the development process. Apache JMeter was used to emulate real-world user activity and assess system responsiveness under different load conditions. To improve testing efficiency and data manipulation, the team implemented a comprehensive automation framework using shell and Perl scripts for remote execution, result collection, and monitoring of key metrics. The structured log outputs from JMeter were analyzed using custom Java programs, which generated detailed reports highlighting application behavior, including response times, CPU usage, and memory consumption. The project further integrated machine learning regression models—Support Vector Regression, AdaBoost, and Gradient Boosting—to predict memory usage and compare model performance during training and testing. Of these, SVR demonstrated superior generalization, while ensemble models, despite