5G Advanced network is known to have substantial energy consumption challenges due to its massive machine-type communication, enhanced mobile broadband and ultra reliable low latency communications. Traditional optimization approach is limited by the high dimensional, dynamic and nonlinear nature of 5G networks. Machine learning is therefore deployed to provide an autonomous, adaptive and predictive energy optimization approach. This study therefore evaluates the performance of three regression models; Linear regression, decision tree regression, and Random Forest regression for predicting energy consumption based on key input features which includes; load, transmission power (TXpower), and energy-saving mode (ESMODE). Model performance was assessed using mean error/mean absolute error (MAE), R-squared values, and visual analysis through scatter plots comparing actual and predicted energy values. The Linear regression model achieved a mean error of 85.91 and an R-squared value of 0.55, indicating limited predictive capability and difficulty in capturing nonlinear relationships, particularly at higher energy levels where underprediction was observed. The Decision tree regression model showed improved performance with a mean absolute error of 41.82 and an R-squared value of 0.78, effectively modeling feature interactions but exhibiting underprediction at high energy values and discrete prediction patterns. The Random Forest Regression model delivered the best results, with a mean absolute error of 36.57 and an R-squared value of 0.81, demonstrating strong predictive accuracy and better generalization across energy ranges. Overall, the results indicate that ensemble-based methods, particularly Random Forest Regression, are more suitable for accurate energy prediction in this context, while simpler linear models are less effective due to their inability to model complex relationships in the data.

5G Decision Trees Machine Learning Regression Random Forest