The need for artificial intelligence (AI) and machine learning (ML) models to optimize data center (DC) operations increases as the volume of operations management data upsurges tremendously. These strategies can assist operators in better understanding their DC operations and help them make informed decisions upfront to maintain service reliability and availability. The strategies include developing models that optimize energy efficiency, identifying inefficient resource utilization and scheduling policies, and predicting outages. In addition to model hyperparameter tuning, feature subset selection (FSS) is critical for identifying relevant features for effectively modeling DC operations to provide insight into the data, optimize model performance, and reduce computational expenses. Hence, this paper introduces the Shapley Additive exPlanation (SHAP) values method, a class of additive feature attribution values for identifying relevant features that is rarely discussed in the literature. We compared its effectiveness with several commonly used, importance-based feature selection methods. The methods were tested on real DC operations data streams obtained from the ENEA CRESCO6 cluster with 20,832 cores. To demonstrate the effectiveness of SHAP compared to other methods, we selected the top ten most important features from each method, retrained the predictive models, and evaluated their performance using the MAE, RMSE, and MPAE evaluation criteria. The results presented in this paper demonstrate that the predictive models trained using features selected with the SHAP-assisted method performed well, with a lower error and a reasonable execution time compared to other methods.

The application of Artificial Intelligence (AI) and Machine Learning (ML) models is increasingly leveraged to automate and optimize Data Centre (DC) operations. However, the interpretability and transparency of these complex models pose critical challenges. Hence, this paper explores the Shapley Additive exPlanations (SHAP) values model explainability method for addressing and enhancing the critical interpretability and transparency challenges of predictive maintenance models. This method computes and assigns Shapley values for each feature, then quantifies and assesses their impact on the model’s output. By quantifying the contribution of each feature, SHAP values can assist DC operators in understanding the underlying reasoning behind the model’s output in order to make proactive decisions. As DC operations are dynamically changing, we additionally investigate how SHAP can capture the temporal behaviors of feature importance in the dynamic DC environment over time. We validate our approach with selected predictive models using an actual dataset from a High-Performance Computing (HPC) DC sourced from the Enea CRESCO6 cluster in Italy. The experimental analyses are formalized using summary, waterfall, force, and dependency explanations. We delve into temporal feature importance analysis to capture the features’ impact on model output over time. The results demonstrate that model explainability can improve model transparency and facilitate collaboration between DC operators and AI systems, which can enhance the operational efficiency and reliability of DCs by providing a quantitative assessment of each feature’s impact on the model’s output.

The need for Artificial Intelligence (AI) and Machine Learning (ML) technologies is increasingly being leveraged for optimizing Data centers’ (DCs’) operations as the volume of operations management data increase tremendously. These strategies can assist operators in better understanding their DC operations and making informed decisions up front to preserve service reliability and availability. Aiming at creating models that optimize energy efficiency, identify inefficient resource utilization and scheduling policies, and predict outages. Apart from model hyperparameter tuning, feature selection is a crucial step to identify relevant features with the objective of providing insight into the data, improving performance, and reducing computational expenses. Although several feature selection methods have been discussed in various domains, none have been discussed in the context of the data center. This paper introduces SHapley Additive exPlanation (SHAP), a class of additive feature attribution values-based feature selection that is rarely discussed in literature. We compared the effectiveness of SHAP method with several widely used methods. We used a real DC dataset obtained from the ENEA CRESCO6 cluster with 2,0832 cores to evaluate the methods. To demonstrate the comparison of the methods, we picked the top 10 most important features from each method, the predictions were retrained, and their performance was evaluated using MAE, RMSE, and MPAE. The results show that the SHAP-assisted feature selection performed best and align with human intuition.

Machine learning (ML) workflows are often analyzed as fixed computational pipelines whose behavior is assumed to remain consistent across execution platforms. However, it remains unclear whether the relative energy contribution of individual stages is an intrinsic property of the workflow or dependent on the execution environment. This work investigates the invariance of stage-level energy contribution patterns across heterogeneous computing environments. We performed operation-level energy measurements of representative ML Data Processing Workflows (DPWs) and reconstruct stage-level energy shares across multiple computational nodes. A pairwise non-parametric analysis reveals statistically significant differences in stage contribution patterns for most node combinations, with limited instances of similarity. In addition, we executed a complete non-synthetic ML data processing workflow to confirm the robustness of our results. A heteroscedasticity-robust two-way ANOVA (HC3 correction) further reveals a significant interaction between pipeline stage and execution environment, demonstrating that platform effects do not merely scale total energy consumption but alter the proportional contribution of stages. The results demonstrate that workflows generally preserve overall cost profiles across stages although their internal composition varies depending on the execution environment. Therefore, ML workflows cannot be characterized by a single stage contribution profile but must be evaluated in the context of specific workflow–environment combinations.