Dikshant Arora completed Machine Learning for Automotive Safety

Built a complete perception-validation dossier for a synthetic Aurora-7 detector: defined detections with IoU >= 0.5, computed aggregate and VRU-specific metrics, and used per-condition recall to surface the night/occluded safety gap. The final recommendation was evidence-based and internally consistent, restricting deployment to daylight conditions until the identified SOTIF-style perception insufficiencies are addressed.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Analysis
• Evidence-Based Deployment Verdict