Energy-Aware Machine Learning: A Comparative Analysis of Predictive Performance and Computational Energy Consumption

Although the increased use of machine learning (ML) has enhanced the accuracy of predictive models in several applications, the computational resources and energy needed by ML algorithms have become important constraints. Past studies have primarily focused on either enhancing the predictive accuracy of ML algorithms or reducing their energy consumption. However, there is research gap in conducting a systematic comparison of the six different ML algorithms in term of predictive performance while considering the computational energy and time consumed in a standard experimental setup. This paper proposes an energy-aware comparative study of six ML algorithms, namely, Logistic Regression, Random Forest, Decision Tree, Support Vector Machine (SVM), XGBoost and Neural Network. The current research will utilize publicly available datasets and standard experimental conditions to evaluate and compare the algorithms based on their prediction performance, computational energy consumption, execution time, CPU utilization and memory utilization. The obtained results will be analysed to establish a relationship between the performance of the ML algorithms and their computational energy consumption to identify the trade-off between accuracy and energy consumption. The current research will provided insight into the need to consider computational costs apart from accuracy while developing machine learning models. The findings of this study can facilitate the creation of energy-efficient and sustainable machine learning models.