How Effective Is Mouse Dynamics Against Web Bots? A Study Across Different Sophistication Levels

Web bots pose an increasing threat to online security with mouse dynamics emerging as a key behavioral biometric for detection. This paper critically evaluates a multi-layered defense strategy against a spectrum of bot attacks from synthetic generation to sophisticated replays. We collected a dataset from two custom websites and evaluated three distinct defensive layers: a suite of machine learning models for behavioral analysis, Dynamic Time Warping (DTW) for historical comparison and click-pattern analysis for contextual validation. Our findings demonstrate that no single technique is a silver-bullet and the optimal defense approach depends on the specific threat model. The supervised Random Forest proved highly effective achieving perfect detection against synthetic attacks and successfully identified the majority of cross-domain replay attacks by leveraging task complexity differences between UI contexts. Furthermore, a simple deterministic click-pattern analysis provided a viable solution against the cross-domain attack highlighting the power of context-aware rules. In contrast, unsupervised models struggled with this context-mismatched replay, while our DTW approach was essential and effective for same-domain replays. This research underscores the necessity of a hybrid, defense-in-depth system that integrates supervised behavioral models, historical analysis and contextual validation to secure mouse-dynamics systems against a comprehensive range of threats.