A Deep Learning Approach to Optimizing Path Loss Prediction in 5G Wireless Networks

Accurate path loss prediction is vital for optimizing 5G wireless network performance, particularly in dense environments. This study investigates the effectiveness of deep learning approaches, specifically Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory Networks (LSTMs), in predicting path loss in 5G. Specifically, key factors influencing path loss prediction, including frequency, distance, and path characteristics, are examined. A comprehensive evaluation using a simulated dataset with varied communication scenarios reveals that RNNs outperform CNNs and LSTMs, achieving an R-squared value of 0.9532, Mean Absolute Error (MAE) of 4.4558, Mean Squared Error (MSE) of 31.3522, and Root Mean Squared Error (RMSE) of 5.5993. These results demonstrate significant improvements in ML-based models over traditional empirical models and highlight the superiority of RNNs in capturing complex temporal dependencies, including weather conditions as input features. This study contributes to understanding deep learning-based path loss prediction, providing valuable insights for optimal design and planning of 5G networks. Comparative analysis with CNNs, RNNs, and LSTMs underscores the importance of model selection in achieving accurate predictions.