Temporal causal feature selection for robust machine learning modelling of data center operations

Modern data centers are complex, energy-intensive systems whose efficient operation is critical for the energy transition. This paper introduces the TempCaFe (Temporal Causal Features) framework, which applies causal feature selection to improve predictive modelling and intervention analysis in data center environments. By integrating causal discovery with domain knowledge for time-series data, the approach not only improves the predictive performance of machine learning models but also their robustness and interpretability. We evaluate the utility of the proposed framework by training and evaluating 4026 models for 31 prediction tasks with data from a data center simulation, a physical testbed, and two operational data centers with distinct cooling architectures (air and water). The results indicate that causal feature selection, compared to traditional feature selection methods, enhances prediction accuracy, especially after system interventions. Furthermore, models trained with causal features require less than one-third of the original features and exhibit greater robustness to data scarcity (number of samples). These findings highlight the potential of causal machine learning to enable robust modelling of data center operations and provide a foundation for decision-making tasks such as what-if analysis and the design of control systems.