🛠️ Predictive Maintenance using Tree-based Models

This project builds a supervised machine learning pipeline to predict equipment failures using historical sensor data.
It includes complete steps from preprocessing to training and evaluation — all in a production-style modular setup.

---

📁 Project Structure

predictive_maintenance/
│
├── data/                 # Raw and processed CSV files
├── notebooks/            # EDA notebook
├── scripts/              # Feature engineering, training, evaluation scripts
├── models/               # Saved models (pkl)
├── output/               # Evaluation outputs (plots & predictions)
├── main.py               # One-click pipeline entry
├── README.md             # Project description
└── requirements.txt      # Required packages

---

🔍 Problem Statement

Can we predict if a machine is going to fail based on sensor readings like torque, temperature, and speed?

This project answers that by:

• Engineering features from raw IoT-style telemetry data
• Training tree-based models (Decision Tree, Random Forest)
• Evaluating performance with accuracy, confusion matrix, and ROC-AUC
• Visualizing model insights (feature importance)

---

📊 Dataset

• Source: AI4I 2020 Predictive Maintenance Dataset
• Size: ~10,000 samples
• Features: Temperature, Speed, Torque, Tool wear, etc.
• Label: Machine failure (0: No, 1: Yes)

---

⚙️ Technologies Used

• Python 3.8
• pandas, NumPy, matplotlib, seaborn
• scikit-learn, joblib
• ROC, AUC, confusion matrix, feature importance

---

🚀 How to Run

1. Clone the repository
2. Create a virtual environment & install requirements:

pip install -r requirements.txt

3. Run the full pipeline:

python main.py

This will:

• Preprocess the data
• Train models
• Generate evaluation outputs (in /output)

---

📈 Results

• Random Forest Accuracy: 99.9%
• Decision Tree Accuracy: 99.8%
• Both models performed well; Random Forest showed better precision with fewer false positives.

🔽 Outputs:

• output/roc_curve.png
• output/feature_importance.png
• output/predictions.csv

---

🧠 Author

Mehmet Ozturk
Feel free to connect: GitHub

---

💬 Feedback

Open to suggestions, ideas, and collaboration!
Follow the repo to stay updated with future ML projects. 🚀