Prediksi Beban Kerja Server Secara Real-Time pada Pusat Data Cloud dengan Pendekatan Gabungan Long Short-Term Memory (LSTM) dan Fuzzy Logic

Efficient resource management in Cloud Data Centers is essential to reduce energy waste and maintain optimal system performance. This study aims to predict server workload in real time using a hybrid approach that combines Long Short-Term Memory (LSTM) and Fuzzy Logic. CPU and RAM usage data were collected every second from a Proxmox Cluster using its API, then normalized and processed using an LSTM model to forecast future workloads. The predicted results were then classified using Fuzzy Logic into three workload categories: light, medium, and heavy. The model was evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), where the results showed an MAE of 2.48 on the training data and 3.09 on the testing data, as well as RMSE values of 5.15 and 5.57, respectively. Based on these evaluation results, the prediction system achieved an accuracy of 97.52% on the training data and 96.91% on the testing data, indicating that the model can generate accurate and stable predictions. This method enables automated decision-making such as workload-based power management, thereby improving energy efficiency and overall system performance.