PREDICTING SELF-LEADERSHIP IN MILITARY CADETS: A MACHINE LEARNING APPROACH

Resumen

Self-leadership is a key competency in military contexts, influencing decision-making, adaptability, and operational performance. This study proposes a data-driven approach to optimise the measurement of self-leadership among cadets at a military academy. Building on the original theoretical framework by Neck and Manz (1992, 1996), and the measurement model developed by Houghton and Neck (2002), which comprises nine factors and 35 items, we applied Recursive Backward Elimination combined with Artificial Neural Networks in MATLAB to iteratively reduce the number of predictive variables. After six elimination rounds, an optimised model with four factors and 18 items was obtained, achieving a predictive accuracy of R² ≈ 0.93 with only a minimal increase in mean squared error. This streamlined version retained 92.6% of the explained variance while significantly reducing the evaluative burden. Comparisons between the original and optimised models, based on classification accuracy and Pearson correlation, revealed high consistency, with the optimised version – hereafter referred to as Mod_4F_RBE – producing the fewest discrepancies. This confirms the robustness of the reduced structure for both continuous and categorical applications. These results highlight the potential of machine learning to optimise psychometric instruments and promote their integration into applied social science. The proposed methodology offers a replicable model for improving assessment efficiency in demanding environments.