December 8, 2025Open Access

A Novel Hybrid Path Loss Prediction Model for 5G Midband Networks Using Empirical, Machine Learning, and Feature Prioritization Techniques

Key Points

Higher prediction accuracy verified with lower mean absolute error of 1.82 dB.
Approach utilizes a hybrid model combining traditional path loss models with machine learning techniques.
Cross-band validation confirms model robustness across different frequencies, with recommendations for fine-tuning.
Enhancing dataset diversity may increase the generalizability of the findings for varied urban environments.

Abstract

Accurate path loss prediction is vital for 5G deployment, especially at midband frequencies where signal degradation is significant. This paper presents a hybrid model that integrates an optimized COST‐231 Hata model with a random forest algorithm to improve prediction accuracy at 3.5 GHz. Recursive feature elimination identified eleven key features from eighteen multidimensional parameters, including novel environmental attributes, to prioritize factors influencing urban path loss. Validation against measurement and simulation datasets showed strong alignment with observed results, achieving lower errors (MAE = 1.82 dB, RMSE = 2.05 dB, and MAPE = 2.4%) compared to existing models. Additionally, cross‐band validation at 1.6 GHz further demonstrated the model’s robustness, though retraining or fine‐tuning is recommended for optimal performance at lower frequencies. Future research may expand the dataset to enhance generalizability.

A Novel Hybrid Path Loss Prediction Model for 5G Midband Networks Using Empirical, Machine Learning, and Feature Prioritization Techniques

Key Points

Abstract

Cite This Study