What Gajendran demonstrated

Built a compact rolling-bearing fault detector end to end: 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 0.39 alarm threshold using the candidate’s own numbers, and the fault call was anchored to the dominant BPFO response identifying an outer-race defect.

• Leakage-Free Feature Engineering
• Honest Model Evaluation
• Embedded-Budget Model Design
• Cost-Based Operational Decisioning