Harshil Patel completed Machine Learning for Automotive Safety

Built a complete perception validation dossier for a synthetic Aurora-7 detector: defined IoU-based detection correctly, scored precision/recall and VRU false-negative rate, and exposed the highest-risk night and occluded slices through per-condition recall analysis. The final recommendation was evidence-based and internally consistent, with SOTIF chains and a real-world Tempe crash comparison tied directly back to the dossier’s weakest operating modes.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Triage
• SOTIF Functional-Insufficiency Reasoning