Exposing the ideal combination of endogenous–exogenous drivers for companies’ ecoinnovative orientation: Results from machine-learning methods

This study provides an XGBoost model to characterize the environmental orientation of innovative firms. This novel approach, using state-of-the-art machine learning methodologies and multiple recognized drivers of eco-innovation, provides solid empirical support for the understanding of the mechanisms that are crucial for firms' transition to a low-carbon economy. Although many drivers have been considered to affect firms’ eco-innovation, our feature selection process using the BorutaShap algorithm demonstrates that few aspects are truly relevant. Furthermore, analyzing a tree surrogate of the final model, our study explores the different paths or combinations of aspects that consistently lead to a specific eco-innovation orientation. The accuracy of the model and the large and complete spectrum of innovative companies in the sample contribute to the generalizability of the results.