Explainable AI Framework for Transparent and Stable Autonomous Traffic Decision-Making

Abstract

More intelligent transportation systems are turning to the use of artificial intelligence to implement adaptive, data-driven traffic control. Although modern machine learning and deep learning models have a significant positive impact on traffic forecasting and signal operation, their black box does not provide transparency and trust in the real-time autonomous decision-making. According to recent research, explainability in traffic systems should be incorporated to enhance reliability and accountability. Nevertheless, the majority of the currently available methods are based on an explanation of post-hoc AI methodologies, which cannot offer consistent explanations in unstable settings. This paper aims to solve these shortcomings by introducing a Context-Aware Explainable Decision Framework (CA-XDF) a unified pipeline that incorporates prediction, explanation generation, temporal consistency validation and confidence-aware decision filtering. The framework uses lightweight machine learning models with real time feature attribution and stability aware mechanism to achieve reliable decision-making. Experimental analysis on real-world traffic data and neuro-simulation based settings show that the proposed framework yields an actuality of 92.6, which is surpassing the baseline models, and enhances the stability of the explanation by about 17 percent and shortening the mean vehicle waiting period to 12 percent. The findings substantiate the importance of installing explainability in the decision loop as it improves interpretability and operational performance. The suggested framework will give a viable and scalable approach to transparent autonomous traffic systems, covering critical issues in describing in real-time decisions