What Muhammad demonstrated

Built a compact rolling-bearing fault detector end to end: visualized the vibration snapshot, engineered time-domain features, trained a shallow decision tree on training bearings only, and evaluated it on a held-out set with catch-rate and false-alarm metrics. The decision memo justified a 0.41 alarm threshold using explicit miss-vs-false-alarm costs and correctly localized the fault to the outer race from the dominant BPFO energy.

• Leakage-Free Feature Engineering
• Honest Model Evaluation
• Cost-Based Decision Making
• Interpreting Defect Frequencies