Berke Ugur Aksakal completed Machine Learning for Automotive Safety

Built a complete perception-validation dossier for a synthetic Aurora-7 detector: defined IoU-based detection correctly, scored aggregate and VRU-specific precision/recall, exposed a severe night-and-occluded VRU recall collapse, and translated the top failures into ISO 21448 SOTIF chains with concrete mitigations. Finished with an internally consistent GO-WITH-CONDITIONS release recommendation grounded in the project’s own numbers and the real Uber Tempe case.

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