Explainable AI helps deep learning predict cloud server loads in real time

A study published in Cluster Computing—by Shabnam Bawa, RajKumar Tekchandani, and Prashant Singh Rana of Thapar Institute of Engineering and Technology in Patiala—targets real-time prediction of cloud server loads using deep temporal forecasting. Released on 27 September 2026, the work reports around 91% predictive performance and the lowest average absolute percentage error among state-of-the-art models compared. The authors say host-level load prediction is crucial for balancing workloads and controlling energy use, since underprovision can cause latency and SLA violations while overprovision wastes electricity. To address challenges from inefficient feature extraction and highly variable workloads, the framework incorporates explainable AI, especially SHAP, to identify feature contributions behind each prediction. The team also builds a real time-series dataset by running multiple containerized applications on virtual machines.