From Correlation to Causation in Lane Change Prediction for Automated Driving: A Causal Explanation Framework

The article introduces a causal-inference framework for lane-change prediction in automated driving, addressing limitations of traditional methods that rely on statistical correlations. This approach integrates deep structural causal modeling, expert-constrained causal discovery, and intervention-based effect analysis, achieving average F1-scores exceeding 95% within three seconds prior to lane changes. This framework enhances interpretability by identifying direct contributors to predictions and their causal relationships, thereby improving decision-making safety in intelligent vehicles.