📊 Defects Prediction System

This project builds a machine learning pipeline to predict the number of product defects in a manufacturing setup using features such as product ID, type, shift, and operator experience. The model is trained using a Random Forest Regressor and provides performance metrics and insightful visualizations.

---

📁 Dataset

Path: ./datasets/aggressive_defects_dataset.csv

📌 Columns:

• Date: Date of production
• Product_ID: Identifier for the product
• Product_Type: Category/type of the product
• Shift: Production shift (e.g., Morning, Evening)
• Operator_Experience_Level: Operator’s experience level
• Defects: Number of defects observed (target variable)

---

🛠 Features

✅ Label encoding for categorical features
✅ Model training using RandomForestRegressor
✅ Evaluation using R² Score and RMSE
✅ Visualizations for feature importance, prediction accuracy, and defect trends
✅ Model & encoders saved using joblib

---

📦 Installation

Install all required dependencies using:

pip install -r requirements.txt

---

🚀 How to Run

1. Place your dataset at:

   ./datasets/aggressive_defects_dataset.csv
   

2. Run the script:

   python defects_predictor.py

3. The script will:

   • Train and evaluate a model
   • Save defect_regressor.pkl and label_encoders.pkl
   • Show feature importance and prediction plots
   • Plot defects trend over time

---

📈 Visualizations

🔹 Top 10 Feature Importances

Shows the most influential features for defect prediction.

🔹 Actual vs Predicted Defects

Compares model predictions with real defect counts.

🔹 Defects Over Time

Line plot to monitor defect trends chronologically.

---

🧪 Model Files

• defect_regressor.pkl: Trained Random Forest model
• label_encoders.pkl: Saved encoders for categorical variables

---

🔍 Predict from CLI

You can later load the model and encoders to predict from new data:

import joblib
import pandas as pd

# Load model and encoders
model = joblib.load("defect_regressor.pkl")
encoders = joblib.load("label_encoders.pkl")

# Sample input (replace with your values)
input_dict = {
    "Product_ID": "P123",
    "Product_Type": "A",
    "Shift": "Night",
    "Operator_Experience_Level":
"Intermediate",
"Machine_usage_hour": 15
}

# Encode input
for col in input_dict:
    le = encoders[col]
    input_dict[col] = le.transform([input_dict[col]])[0]

# Predict
X_input = pd.DataFrame([input_dict])
predicted_defects = model.predict(X_input)[0]
print(f"Predicted Defects: {predicted_defects:.2f}")

---

📤 Future Improvements

• Add a web dashboard using Flask or Streamlit
• Hyperparameter tuning with GridSearchCV
• Integration with real-time factory data sources
• Support for more advanced models (XGBoost, CatBoost)

---

👨‍💻 Author

Raj Aryan 🎓 B.Tech | RNSIT 🔗 LinkedIn 🔗 GitHub

---

📝 License

This project is open-source and free to use under the MIT License.