youssef badaoui completed Machine Learning for Automotive Safety

Built an end-to-end perception validation dossier for a fictional Aurora-7 detector: defined IoU-based detection correctly, computed aggregate precision/recall and VRU false-negative rate, sliced recall by lighting/occlusion/class to expose the dangerous night-time and occluded pedestrian failures, and translated those misses into ISO 21448 SOTIF safety cases with concrete mitigations. Finished with an evidence-based daylight-restricted deployment recommendation tied back to the metrics and the Uber Tempe crash analysis.

• Leakage-Free Metric Definition
• Condition-Sliced Safety Analysis
• SOTIF Functional-Insufficiency Reasoning
• Evidence-Based Deployment Decisioning