Modeling Organizational Ambidexterity through Ensemble Learning: Behavioral and Structural Predictors of Exploratory and Exploitative Innovation

Abstract

Objective: The objective of this study was to model organizational ambidexterity by applying ensemble machine learning techniques to identify the behavioral and structural predictors of exploratory and exploitative innovation.

Methods and Materials: This explanatory study employed a cross-sectional design involving 487 middle- and senior-level managers from medium and large organizations across major industries in Chile. Data were collected using validated instruments measuring leadership cognitive flexibility, learning orientation, psychological safety, risk tolerance, cross-functional integration, decentralization, resource flexibility, knowledge-sharing systems, and dual innovation outcomes. The analytical framework integrated traditional statistical validation with an ensemble learning architecture composed of Random Forest, Gradient Boosting, XGBoost, and Support Vector Regression models. Model training applied stratified sampling, five-fold cross-validation, and hyperparameter optimization, while performance was evaluated using R², RMSE, MAE, and explained variance. Explainable AI techniques based on SHAP were employed to interpret nonlinear relationships and predictor contributions.

Findings: The ensemble model demonstrated superior predictive performance for both exploratory innovation (R² = 0.81, RMSE = 0.25) and exploitative innovation (R² = 0.84, RMSE = 0.22), significantly outperforming individual machine learning algorithms. Leadership cognitive flexibility and learning orientation emerged as the strongest predictors of exploratory innovation, whereas cross-functional integration and structural decentralization exerted the greatest influence on exploitative innovation. Psychological safety, risk tolerance, knowledge sharing, and resource flexibility contributed significantly to both innovation dimensions, with SHAP analysis revealing asymmetric and nonlinear interaction effects across predictors.

Conclusion: The results confirm that organizational ambidexterity is a systemic, nonlinear phenomenon driven by the dynamic interaction of behavioral and structural factors and that ensemble learning provides a powerful methodological approach for modeling this complexity, offering both theoretical advancement and practical guidance for innovation management.