predictive-maintenance-model

Predictive Maintenance Model - Engine Failure Prediction

Model Description

This model predicts engine failures for automotive predictive maintenance using sensor data.

Model Type: AdaBoost Task: Binary Classification (Normal vs Faulty Engine) Framework: scikit-learn / XGBoost

Model Performance

Test Set Metrics

• Accuracy: 0.6668
• Precision: 0.6854
• Recall: 0.8713 (Primary metric - minimizes false negatives)
• F1-Score: 0.7673
• ROC-AUC: 0.6959

Model Details

Hyperparameters

{
  "learning_rate": 0.05,
  "n_estimators": 100
}

Training Information

• Training Samples: 15,628
• Test Samples: 3,907
• Features: 17
• Training Date: 2026-02-08 16:01:01

Features

The model uses 17 features including:

• Engine RPM
• Lubricating oil pressure and temperature
• Fuel pressure
• Coolant pressure and temperature
• Engineered features (temperature-pressure ratios, differentials, etc.)

Usage

import joblib
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(
    repo_id="SharleyK/predictive-maintenance-model",
    filename="best_model.pkl"
)

# Load model
model = joblib.load(model_path)

# Download scaler
scaler_path = hf_hub_download(
    repo_id="SharleyK/predictive-maintenance-model",
    filename="scaler.pkl"
)
scaler = joblib.load(scaler_path)

# Make predictions
X_new_scaled = scaler.transform(X_new)
predictions = model.predict(X_new_scaled)
probabilities = model.predict_proba(X_new_scaled)

# Interpret results
# 0 = Normal/Healthy Engine
# 1 = Faulty/Requires Maintenance

Model Selection

This model was selected from 6 candidates:

• Decision Tree
• Bagging Classifier
• Random Forest
• AdaBoost
• Gradient Boosting
• XGBoost

Selection criteria: Highest test recall (to minimize false negatives - missed failures)

Business Impact

• Reduces unplanned breakdowns by detecting failures early
• Minimizes emergency repair costs
• Optimizes maintenance scheduling
• Improves fleet availability and safety

Limitations

• Requires all sensor inputs to be available
• Trained on specific engine types (automotive and small engines)
• Performance may degrade if sensor calibration changes
• Requires periodic retraining with new data

Citation

@model{predictive_maintenance_engine_model,
  author = {SharleyK},
  title = {Predictive Maintenance Model - Engine Failure Prediction},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/SharleyK/predictive-maintenance-model}
}

License

MIT License