Multiple Linear Regression Analysis: Predicting Highway MPG

Project Overview

This project applies Multiple Linear Regression (MLR) to predict Highway mpg (miles per gallon) of a car(Fuel Information.Highway MPG) based on various vehicle features. The dataset includes technical specifications of cars, such as engine type, fuel type, and transmission details.

---

Objective

• Identify key factors affecting highway fuel efficiency.
• Build a robust regression model to predict Highway mpg (miles per gallon) of a car(Fuel Information.Highway MPG).
• Evaluate the model’s performance and optimize feature selection.

---

Project Structure

📂 MLR Analysis/
│
├── 📄 LICENSE
│
├── 📄 README.md
│
├── 📂 data/
│   └── cars.csv  # Your dataset
│
└── 📄 MLR_analysis.ipynb  # Google Colab notebook with the full analysis

Dataset Information

The dataset includes the following features:

• Vehicle Specifications: Identification.Model Year, Dimensions.Height, Dimensions.Width, etc.
• Engine Details: Engine Information.Engine Type, Torque, Horsepower, etc.
• Fuel Information: Fuel Type, City MPG, etc.
• Transmission Details: Number of Forward Gears, Transmission Type.

---

Data Preprocessing Steps

• Handling duplicated Values: Removed duplicated values
• Encoding Categorical Variables:
  • Applied Target Encoding for 'Identification.Make', 'Identification.Model Year', 'Engine Information.Engine Type','Engine Information.Driveline'
  • Applied One-Hot Encoding for 'Fuel Information.Fuel Type','Engine Information.Transmission','Identification.Classification'
• Feature Scaling: Standardized numerical features where necessary.
• Outlier Treatment: Used IQR method to remove extreme values.

---

Exploratory Data Analysis (EDA)

Correlation Heatmap

A heatmap was generated to analyze the correlation between independent variables and the target variable (Fuel Information.Highway MPG).

Key Observations:

• Fuel Information.City mpg is highly correlated with Highway MPG.
• Some features exhibit multicollinearity, requiring feature selection.

---

Feature Selection

To select the best predictors:

• Recursive Feature Elimination (RFE): Selected the most impactful features.
• Multicollinearity Check (VIF): Removed features with VIF > 10 to avoid redundancy.
• Correlation Analysis: Chose features highly correlated with the target variable but uncorrelated with each other.

Final Selected Features:

['Fuel Information.City mpg', 'Identification.Model Year_encoded', 
 'Fuel Information.Fuel Type_Diesel fuel', 'Fuel Information.Fuel Type_E85', 
 'Fuel Information.Fuel Type_Gasoline']

---

Model Building

• Model Used: Multiple Linear Regression
• Train-Test Split: 80% Training, 20% Testing
• Evaluation Metrics:
  • R² Score: Indicates how well the model explains variance.
  • VIF Analysis: Ensures low multicollinearity.
  • RMSE: Measures average error in predictions.

---

Original vs Predicted data

Here’s the comparison between the actual values and the predicted values:

Results & Model Evaluation

Metric	Before Handling Outliers	After Handling Outliers
R² Score	0.9121	0.9495
RMSE	0.0070	0.0432
Adjusted R²	0.9357	0.9529

After handling outliers, the model became more generalizable with reduced errors.

---

Residual Analysis

To check model assumptions, we plotted residuals to ensure they followed a normal distribution and exhibited homoscedasticity.

Key Takeaways from Residual Analysis:

• The residuals approximately follow a normal distribution.
• No clear heteroscedasticity, indicating stable variance.
• Confirms that our model meets linear regression assumptions.

---

Key Takeaways

• Feature Engineering Matters: Proper encoding and selection significantly improved model performance.
• Multicollinearity is Crucial: Reducing VIF led to more stable coefficients.
• Outlier Handling is Important: Post-cleaning, the model showed better predictive accuracy.
• Business Impact: This model helps automobile manufacturers understand which factors most influence fuel efficiency.

---

How to Run Your Code

Open the Google Colab Notebook

Click the link below to open the project in Google Colab:

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Aparna-analyst/Machine-Learning/blob/main/MLR_analysis.ipynb)

---

Upload the Dataset

• Make sure to upload your dataset (cars.csv) to Colab by:
  • Clicking Files on the left sidebar.
  • Clicking Upload and selecting your CSV file.

---

Install Dependencies in Colab

If needed, install dependencies directly in Colab by running:

!pip install pandas numpy scikit-learn matplotlib seaborn

---

Run the Code

• Execute all the cells in the notebook step by step by pressing Shift + Enter.

---

Save Results

• Download your output files by right-clicking on them in the Files section and choosing Download.

---

Next Steps

• Try Ridge/Lasso Regression to improve regularization.
• Perform Cross-Validation for better generalization.
• Test on new unseen vehicle data to validate real-world performance.

---

References

• Scikit-Learn Documentation: https://scikit-learn.org/
• Pandas Data Manipulation: https://pandas.pydata.org/

---

Contributors

Aparna S - LinkedIn

---