🚗 Auto MPG Regression – Machine Learning Project

This project focuses on predicting the fuel efficiency (MPG - Miles Per Gallon) of automobiles using their technical specifications. It applies various data preprocessing techniques, regression algorithms, and ensemble methods to develop accurate and robust predictive models.

---

📂 Dataset

• Source: UCI Machine Learning Repository – Auto MPG Dataset
• Total Instances: 398
• Features Used:
  • Cylinders
  • Displacement
  • Horsepower
  • Weight
  • Acceleration
  • Model Year
  • Origin
• Target Variable: MPG (Miles Per Gallon)

---

📊 Workflow Overview

1. Data Cleaning
   • Handled 12 missing values in the Horsepower column using group-wise median imputation (based on Cylinder count).
2. Exploratory Data Analysis (EDA)
   • Correlation matrix, clustermap, skewness analysis, and distribution plots.
3. Outlier Removal
   • Used IQR method on Horsepower and Acceleration.
4. Feature Engineering
   • Log transformation applied to target (MPG) due to skewness.
   • One-hot encoding on Cylinders and Origin.
5. Data Splitting & Scaling
   • 10% training / 90% testing split to challenge the models.
   • Applied RobustScaler to minimize the influence of outliers.
6. Modeling
   • Applied and tuned:
     • Linear Regression
     • Ridge Regression (L2)
     • Lasso Regression (L1)
     • ElasticNet (L1 + L2)
     • XGBoost Regressor
     • Averaging Ensemble (Lasso + XGBoost)
7. Performance Evaluation
   • Mean Squared Error (MSE) used as the evaluation metric.

---

🧠 Model Comparison (MSE on Test Set)

Model	MSE	Notes
Linear Regression	0.01363	Baseline model
Ridge Regression	0.01340	Slightly improved with L2 regularization
Lasso Regression	0.01331 ✅	Feature selection + low error
ElasticNet	0.01330 ✅✅	Best performance, balanced model
XGBoost	0.01810 ❌	Overcomplicated for this small dataset
Averaged (Lasso + XGB)	0.01365	Balanced but not outperforming Lasso/ENet

---

🧪 Key Insights

• ElasticNet and Lasso were the most effective models for this dataset.
• RobustScaler proved useful in stabilizing model performance.
• XGBoost, while powerful, underperformed due to dataset simplicity.
• Averaging models yielded consistent but not superior results.

---

📦 Technologies Used

• Python 3.10+
• NumPy, Pandas
• Seaborn, Matplotlib
• Scikit-learn
• XGBoost
• Jupyter Notebook / VSCode

---

📁 Project Structure