# Predictive Maintenance — Engine Failure Detection

## Model Details
- **Best Model**     : Random Forest
- **Task**           : Binary Classification (0=Normal, 1=Faulty)
- **Dataset**        : abhisekbasu/predictive-maintenance-engine
- **Training Records**: 15,628
- **Test Records**   : 3,907
- **Primary Metric** : Recall (Faulty class)

## Performance — Best Model (Random Forest)
| Metric | Score |
|--------|-------|
| Accuracy | 0.6473 |
| Precision | 0.6862 |
| Recall | 0.8116 |
| F1-Score | 0.7437 |
| ROC-AUC | 0.6750 |

## All Models Comparison
| Model | Accuracy | Precision | Recall | F1 | ROC-AUC |
|-------|----------|-----------|--------|----|---------|
| Decision Tree | 0.5923 | 0.6789 | 0.6703 | 0.6746 | 0.5647 |
| Random Forest | 0.6473 | 0.6862 | 0.8116 | 0.7437 | 0.6750 |
| XGBoost | 0.6222 | 0.7052 | 0.6886 | 0.6968 | 0.6563 |

## Features Used
- **Original** : Engine_RPM, Lub_Oil_Pressure, Fuel_Pressure,
  Coolant_Pressure, Lub_Oil_Temperature, Coolant_Temperature
- **Engineered** : Pressure_Index, Thermal_Index,
  Pressure_Temp_Ratio, RPM_Pressure_Ratio, Thermal_Deviation

## Files
| File | Description |
|------|-------------|
| best_engine_model_v1.joblib | Trained Random Forest model |
| model_summary.json | Complete metrics and parameters |
| mlruns/mlruns.zip | All MLflow experiment logs |

## Experiment Tracking
All hyperparameter combinations were tracked using MLflow.
Complete experiment logs are available in mlruns/mlruns.zip.