Explainable AI with EDA for V2I path loss prediction.

Area of Science:

• Wireless Communications
• Machine Learning
• Signal Propagation

Background:

• Accurate pathloss (PL) prediction is critical for reliable Vehicle-to-Infrastructure (V2I) communication in complex urban environments.
• Traditional empirical models and black-box machine learning (ML) methods have limitations in accuracy, transparency, and suitability for safety-critical V2X applications.

Purpose of the Study:

• To propose a fully explainable V2I PL prediction framework.
• To enhance transparency and trustworthiness in V2X communication systems.
• To provide robust global and local explanations of feature contributions in PL prediction.

Main Methods:

• Integration of Exploratory Data Analysis (EDA), optimized Kalman filtering, and interpretable ML models (EBM, GAM, GNAM).
• Validation using a large-scale dataset across 24 heterogeneous urban scenarios.
• Evaluation through 5-fold cross-validation and multi-seed runs.

Main Results:

• Interpretable models demonstrate competitive accuracy compared to black-box approaches.
• The framework provides robust global and local explanations of feature contributions.
• The proposed models are computationally feasible for real-time V2X deployment.

Conclusions:

• The explainable V2I PL prediction framework offers a transparent and trustworthy solution for future 5G/6G systems.
• Interpretable ML models are suitable for safety-critical V2X applications requiring explainability.
• The study addresses computational, real-time, and ethical considerations for practical V2X deployment.