A Predictive Ensemble Modeling Approach Linking Psychological Doubt to Code Refactoring and Productivity Losses

• Adaan Ahsun

  Covenant University

  Author

Software development organizations face persistent challenges in predicting and mitigating productivity losses, yet the psychological antecedents of developer behavior remain underexplored in quantitative models. While research has established that technical debt impedes productivity and that self-efficacy shapes developer performance, no validated framework exists that links psychological doubt to specific refactoring behaviors and downstream productivity outcomes. This study develops and validates a predictive ensemble modeling approach that treats psychological doubt as a leading indicator of excessive code refactoring and associated productivity losses. Drawing on the OSMI Mental Health in Tech Survey (n=1,428) integrated with simulated refactoring frequency data grounded in established maintainability research, we compare Random Forest, XGBoost, Support Vector Machine, and Logistic Regression models. The Random Forest ensemble demonstrated superior performance with 89.4% prediction accuracy (AUC = 0.912, precision = 0.887, recall = 0.903), significantly outperforming baseline approaches (p < .001). Feature importance analysis identified doubt-related variables—including imposter syndrome indicators and self-efficacy deficits—as the top predictors, accounting for 37.2% of model variance. The framework offers practitioners an actionable tool for early identification of doubt-driven refactoring cycles, enabling targeted interventions that may reduce productivity losses by an estimated 18–24%. This research contributes a replicable methodology for integrating psychological constructs into software engineering predictive analytics.

• pdf

Copyright (c) 2026 Adaan Ahsun (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.