Identifying Opponent's Neuroticism Based on Behavior in Wargame

Traditional neuroticism assessments primarily rely on self-report questionnaires, which can be difficult to implement in highly confrontational scenarios and are susceptible to subjective biases. To overcome these limitations, this study develops a machine learning-based approach using behavioral data to predict an opponent's neuroticism in competitive environments. We analyzed behavioral records from 167 participants on the MiaoSuan Wargame platform. After data cleaning and feature selection, key behavioral features associated with neuroticism were identified, and predictive models were developed. Neuroticism was assessed using the 8-item neuroticism subscale of the Big Five Inventory. Results indicate that this method can effectively infer an individual's neuroticism level. The best-performing model was LinearSVR, which balances interpretability, robustness to noise, and the ability to capture moderate nonlinear relationships-making it suitable for behavior-based psychological inference tasks. The correlation between predicted scores and self-reported questionnaire scores was 0.606, the R-squared value was 0.354, and the test-retest reliability was 0.516. These behavioral features provide valuable insights into neuroticism prediction and have practical applications in psychological assessment, particularly in competitive environments where conventional methods are impractical. This study demonstrates the feasibility of behavior-based neuroticism assessment and suggests future research directions, including refining feature selection techniques and expanding the application scenarios.