Machine Utilization Prediction in Cloud Using Informer Model

This study investigates CPU utilization prediction in cloud environments using the Informer model, a deep learning framework optimized for long-sequence forecasting. By leveraging its ability to capture complex temporal patterns, the Informer model is benchmarked against GRU, LSTM, Prophet, and XGBoost across diverse CPU utilization datasets. Performance is evaluated using RMSE, MSE, MAE, and MAPE metrics. Experimental results demonstrate that the Informer model significantly outperforms the comparison models, excelling in capturing both gradual trends and abrupt fluctuations. Specifically, the Informer achieved a 47.51% reduction in RMSE over GRU, 49.90% over LSTM, 67.36% over Prophet, and 50.46% over XGBoost. These findings underscore the Informer model's superior adaptability and precision, making it a robust solution for dynamic resource utilization prediction in cloud environments. Its predictive accuracy is pivotal for enhancing workload management and optimizing resource allocation.