Mechanical Engineering: Predictive Maintenance & Vibration

At a glance

Task: classification  |  Headline: accuracy = 0.9924

Summary

Project

Config-driven data science pipeline for high-fidelity industrial domains.

What was investigated

The engagement modelled the target 'failure_type' as a classification problem in the Mechanical Engineering: Predictive Maintenance & Vibration domain, using 2000 records across 19 columns (data source: Synthetic sandbox). The pipeline audited data quality, engineered features, split the data honestly into train/test, and compared several models by cross-validation.

Outcome

The selected model was LightGBM, with f1_macro = 0.9093 on the held-out test set. Supporting metrics: accuracy=0.992, balanced_accuracy=0.888, precision_weighted=0.993, recall_weighted=0.992, f1_weighted=0.991, precision_macro=0.986, recall_macro=0.888, cohen_kappa=0.982, matthews_. Full results, the deployment gates and the audit trail are in the run's report.

Figures

Artifacts

Data source: Synthetic sandbox

Audit SHA-256: 69732d1fef320a9ba2074c9e5f475346be29aff9b644ece578df88e30a066a46