Ŷ UGO: Enhancing Power Consumption Predictions in Data Centers Through Combined Multiple Linear Regression and Gradient Boosting Techniques on Hypervisor Load
Yuggo Afrianto, Rendy Munadi, Setyorini, Toto Widyanto · IEEE Access · 2025
Accurate power consumption prediction in data centers is critical for optimizing energy efficiency, but is fundamentally challenged by workloads exhibiting both linear and non-linear patterns. This study proposes ŶUGO model, a novel two-stage hybrid model that integrates Gradient Boosting (GB) for non-linear workload forecasting and Multiple Linear Regression (MLR) for final power prediction, leveraging proxy metrics from hypervisor loads via SNMP. Through extensive experimentation, ŶUGO model demonstrates superior accuracy, achieving a Mean Absolute Error (MAE) of 1.46 and a Root Mean Squared Error (RMSE) of 1.50 in multi-step-ahead forecasting. Crucially, this performance represents a significant MAE reduction of up to 86.42% compared to the ARIMA model and substantial improvements over MLP and SVR benchmarks. The proposed framework offers a robust, scalable, and cost-effective solution for precise energy monitoring without requiring physical sensors, providing a practical tool for enhanced resource management in virtualized environments.