Deep Neuro-Fuzzy Capability Analysis for Reliable Autonomous Vehicle Communications

Abstract

Autonomous Vehicles (AVs) rely on high-assurance communication units to maintain operational safety; however, approximately 40% of communication failures are attributed to inconsistent unit performance, and 35% of real-time decision conflicts stem from network latency. Conventional evaluation methodologies are often reactive, failing to capture the subtle, non-linear performance variations inherent in dynamic traffic environments. To address these systemic vulnerabilities, this research proposes a Deep Fuzzy Logic (DFL)-based framework for the automated capability assessment of AV communication units. The methodology introduces a Deep Fuzzy Encoding (DFE) layer designed to transform raw network metrics into high-dimensional fuzzy representations, effectively modeling the stochastic uncertainty of signal integrity and latency. Following rigorous preprocessing—including outlier elimination and feature normalization—the DFE-extracted features are processed via a Linear Regression architecture to estimate a continuous capability score. Comparative analysis against baseline Decision Tree Regressor (DTR) and K-Nearest Neighbor (KNN) regressors demonstrates that the integration of fuzziness significantly enhances prediction reliability and error convergence. The framework is deployed as a Flask-based web application, providing a scalable, real-time diagnostic tool for proactive maintenance and the optimization of autonomous vehicular networks.