Mouse Dynamics Dataset for Behavioral User Substitution Detection in Electronic Testing

This dataset contains streaming behavioral (mouse-interaction) events recorded during computer-based testing sessions between 2015 and 2018. It comprises 31,990,875 raw events collected from 1,755 users, including 9.48 million click events and 22.51 million mouse-movement events. All records are fully anonymized: no names, identifiers, IP addresses, or any personally identifiable information are included. Each user is represented only by an internal numeric ID with no link to real-world identity. Each event record includes: event type (click or movement); click duration (ms); movement speed (pixels/second); movement length (pixels); movement direction (angle); and inter-event timing used to derive tempo features. Data are provided in raw form and, separately, after removal of physically implausible outliers (2.03% of records), such as negative or extreme speed values and unrealistic click durations. For research reproducibility, the dataset also includes intermediate artifacts used in the associated study: cleaned event tables, non-overlapping windowed behavioral profiles (window sizes W = 50, 100, 200 events) built from directional-kinematic features (quantile signatures of click duration, movement speed and length; circular statistics of movement direction; inter-event tempo; click share), and per-user reference/probe window splits used for verification experiments. A refined cohort of 1,067 users meeting minimum activity thresholds (at least 3,000 events, 5 sessions, 200 clicks, and 1,000 movements) is identified via a cohort-selection flag. The dataset is intended to support research on behavioral biometrics, mouse dynamics, continuous and open-set user authentication, anomaly and drift detection in streaming data, and user-substitution detection in remote/e-testing and similar human-computer interaction contexts. It enables reproducible evaluation of distributional and sequential-monitoring methods (e.g., Wasserstein-distance-based criteria, CUSUM-type detectors) without requiring access to any personal or identifying information. Data are shared as CSV.