Predictive Maintenance – Gradient Boosting Model
Model Overview
This model is a recall-optimized Gradient Boosting classifier developed to support predictive maintenance for engine systems. The primary objective is to identify engines likely to require maintenance before failure occurs.
Training Data
The model was trained on a prepared engine sensor dataset sourced from the Hugging Face Dataset Hub. The dataset contains structured numeric sensor readings representing engine operating conditions.
Objective
- Minimize missed engine failures (false negatives)
- Prioritize recall for the faulty engine class
Evaluation Metrics
- Recall (Faulty): ~0.84
- ROC-AUC: ~0.70
- PR-AUC: ~0.80
Intended Use
This model is intended for:
- Predictive maintenance decision support
- Risk-based maintenance scheduling
- Offline or batch inference scenarios
Limitations
- Trained on a static, pre-processed dataset
- Does not incorporate temporal or sequential dependencies
- Threshold selection may require calibration based on operational risk tolerance
Model Artifacts
The repository contains a serialized joblib model file that can be loaded directly for inference in Python-based environments.
- Downloads last month
- -