premswan/engine-predictive-maintenance-model

Engine Predictive Maintenance Model

Business Objective

This model predicts whether an engine is operating normally or requires maintenance based on historical engine sensor readings.

Dataset

Training and test data were loaded from the Hugging Face dataset repository:

premswan/engine-predictive-maintenance-data

Input Features

• Engine_RPM
• Lub_Oil_Pressure
• Fuel_Pressure
• Coolant_Pressure
• Lub_Oil_Temperature
• Coolant_Temperature

Target Variable

Engine_Condition

• 0: Normal / operating condition
• 1: Faulty / requires maintenance

Model Selection

Best model selected by test F1 score:

GradientBoosting

Best Hyperparameters

{
  "model__subsample": 0.8,
  "model__n_estimators": 100,
  "model__max_depth": 5,
  "model__learning_rate": 0.01
}

Test Metrics

Metric	Value
Accuracy	0.6560
Precision	0.6582
Recall	0.9452
F1 Score	0.7760
ROC AUC	0.6976

Intended Use

This model is intended for predictive maintenance analysis using engine sensor data. It can support early identification of potential engine issues and help plan proactive maintenance.

Limitations

• Model performance depends on the representativeness and quality of the training data.
• Sensor drift, new engine types, or changed operating conditions can reduce model reliability.
• The model should be monitored and periodically retrained with newer production data.

Artifacts

• best_model.joblib: Trained scikit-learn pipeline
• metrics.json: Test metrics
• best_params.json: Tuned hyperparameters
• feature_list.json: Required input feature list
• label_map.json: Target label interpretation
• experiment_summary.csv: Cross-validation experiment summary
• model_evaluation_metrics.csv: Test-set metrics for all tuned models
• feature_importance.csv: Feature importance, if supported by the selected model