A Lightweight Machine Learning Framework for Predicting User Frustration from Interaction Logs to Enhance UX/UI Design

This research presents a lightweight machine learning framework designed to automatically predict user frustration by analyzing interaction logs collected during digital product usage. The study addresses a major gap in current UX/UI design methodologies: the lack of fast, automated, and cost-effective tools that can detect user frustration without manual usability testing. Using BagAChat.com as a case study, synthetic Hotjar-style event logs were generated to model real user interactions including clicks, idle time, scroll behavior, errors, navigation loops, and rage clicks. A comprehensive preprocessing and feature engineering pipeline converted noisy event streams into structured behavioral features. Several ML models were evaluated, with Random Forest achieving the highest predictive performance. The resulting framework is computationally efficient, interpretable, and suitable for integration into real-time UX monitoring dashboards. Alongside the research paper, a functional Streamlit demonstration is provided to illustrate how interaction logs can be uploaded and analyzed for frustration prediction. This submission contains the full research paper, dataset description, methodology, results, conclusion, and future work. It contributes a practical solution for modern UX/UI teams seeking data-driven design improvements.