Exploring the Impact of K-Anonymisation on the Energy Efficiency of Machine Learning Algorithms

With the increased use of Artificial Intelligence (AI), concerns about AI's energy consumption are increasing as well. This paper investigates the impact of k-anonymisation and dataset characteristics on energy consumption during machine learning (ML) training. U sing three datasets from the UCI Machine Learning Repository, we analyze the energy efficiency of ML algorithms-Random Forest (RF), k-Nearest neighbours (KNN), and Logistic Regression (LR)-trained on both k-anonymised and original datasets. Our experiment reveals that k-anonymisation significantly reduces energy consumption during Random Forest (RF) and Logistic Regression (LR) training. Additionally, we find that k-anonymisation leads to greater energy savings in Logistic Regression (LR) training if more features are present in the dataset. However, we also find that the energy savings do not hold in the KNN case, except for one feature case. These findings are backed by Aligned Ranked Transform Analysis of Variance on empirically measured energy consumption data. Our work strengthens the need for further empirical exploration into energy efficiency in ML algorithms amidst the growing demand for AI.