Predicting leadership emergence in global virtual teams: machine learning applications in algorithmic HRM

Abstract

Identifying leadership talent early remains a persistent challenge for organizations, particularly in distributed and multicultural settings. Responding to calls for predictive approaches in algorithmic HRM, this study evaluates the relative performance and interpretability of supervised machine learning models for predicting leadership emergence in non-hierarchical global virtual teams (GVTs). Using data from 8,112 individuals nested in 2,371 GVTs, we benchmark five supervised algorithms (Artificial Neural Networks, Naïve Bayes, Random Forest, XGBoost, and Logistic Regression) on out-of-sample classification performance using demographic, psychological, motivational, behavioral, and team climate features collected across multiple time points. All models demonstrate strong predictive validity, with XGBoost achieving modestly superior overall performance (79% accuracy; AUROC = 0.87), consistent with its ability to capture nonlinear relationships in complex, high-dimensional data. Explainable AI analyses using SHAP identify the features that contribute most strongly to leadership emergence predictions (help with coordination, effort, friendliness (inverse), English language proficiency, and perceived team efficacy (inverse)) and reveal nonlinear predictive patterns that are difficult to detect with traditional statistical approaches. By framing leadership identification as a predictive talent analytics problem, this study advances algorithmic HRM research while illustrating how interpretable machine learning models can generate theoretically informative insights for leadership research in global and virtual contexts.