Green AI in the Finance Industry: Exploring the Impact of Feature Engineering on the Accuracy and Computational Time of Machine Learning Models

As research and practice on applications of Artificial Intelligence (AI) exponentially increase, the support for deployment grows at the same rate. While a large amount of data available enables sophisticated methods to perform feature engineering, reaching higher accuracy, it is imperative to emphasize the computational costs and the efficiency level in which these models operate. This paper contrasts the processing time and accuracy of individual and hybrid Machine Learning (ML) models obtained when predicting customer loyalty in financial settings. We use frameworks that account for feature engineering and green AI philosophy aspects separately within the individual and hybrid proposed approaches. The individual models refer to commonly used regressor-based algorithms (e.g., decision trees, gradient boosting, and LightGBM) widely applied in business problems. The hybrid models use k-Means to cluster customers before implementing the individual regressor-based models. Our findings indicate that using a lower number of features results in a slightly smaller accuracy than models incorporating features. Besides, we explicitly illustrate the trade-off between the higher accuracy and computational time of the hybrid ML models against the lower accuracy and computational time of the individual models when assessing customers' loyalty levels. Thus, our results provide managers with information regarding the model to be deployed based on their firms' specifications.