An explainable deep temporal framework for cloud-based real-time load forecasting

doi:10.1007/s10586-026-06599-4

Return

An explainable deep temporal framework for cloud-based real-time load forecasting

2026-09-27

PRE

Shabnam Bawa *

RajKumar Tekchandani

Prashant Singh Rana

DOI：10.1007/s10586-026-06599-4

Original

Original request for help

Save

Abstract

En 中文

In order to optimize resource utilization, balance load and minimize energy consumption, load prediction on the host is indispensable in cloud computing. However, inefficiency in feature extraction and variations in load constitute an important obstacle to prediction of host load. To improve workload prediction, this paper proposes a novel explainable deep time-series-based framework for real-time load forecasting in cloud environment, incorporating explainability techniques (XAI) such as SHapley Additive exPlanations (SHAP). Moreover, a real time series based dataset is generated while running multiple applications in the form of containers on virtual machines. The performance of the proposed model is evaluated against several deep learning approaches such as CNN-LSTM, GRU, TCN, Seq-to-Seq, Autoencoder and GNN for forecasting time series data using metrics such as accuracy, MAPE, RMSE and MSE. Further, XAI techniques are used for feature importance analysis. In this work, SHAP outperforms due to its consistent and transparent evaluations. For the purpose of experimentation, the proposed model’s efficacy is evaluated in comparison to other existing state of the art approaches. Compared to current cutting-edge models, the proposed model demonstrates a higher level of accuracy in the prediction of workload. It obtains the lowest average absolute percentage error and a predictive performance of approximately 91%.

Keywords:

Load prediction

Containers

Deep learning

Explainable-AI

Cloud computing