Gradient_Descent_MLR_Implementation

Gradient Descent implementation for Multiple linear regression

---

📊 Multiple Linear Regression from Scratch (Gradient Descent)

This project demonstrates a basic implementation of Multiple Linear Regression (MLR) using Gradient Descent, without using sklearn's regression models. It's designed for educational purposes to show how linear models work under the hood.

---

🚀 Features

• Implements multiple linear regression using gradient descent.
• Scales features using StandardScaler for better convergence.
• Custom training (fit) and prediction (predict) methods.
• Works on any dataset with a numeric target column named income.

---

🧠 Algorithm

The model optimizes weights using the gradient descent algorithm:

1. Initialize weights to zero.
2. Scale features using Z-score normalization.
3. Iteratively update weights using the gradient of the Mean Squared Error (MSE) loss function.

---

🛠️ Dependencies

• numpy
• pandas
• scikit-learn (for StandardScaler)

Install dependencies with:

pip install numpy pandas scikit-learn

---

📁 Dataset

The dataset should be a CSV file containing multiple numerical features and a target column named income.

Example path:

Multiple Linear Regression/multiple_linear_regression_dataset.csv

---

📌 Usage

import pandas as pd
from mlr import MLR_Grad_Desc  # Assuming you've saved the class in a file named mlr.py

# Load data
data = pd.read_csv("Multiple Linear Regression/multiple_linear_regression_dataset.csv")

# Initialize and train the model
mlr = MLR_Grad_Desc(data, lr=0.01)
mlr.fit(epochs=300)

# View learned weights
print(mlr.wgt)

# Predict on new data
new_data = pd.DataFrame({...})  # Replace with actual feature data
predictions = mlr.predict(new_data)

---

📈 Output

The final weights learned by the model will be printed after training:

[weight_0, weight_1, weight_2, ..., weight_n]

---

⚠️ Notes

• Make sure the income column is present in your dataset.
• Feature scaling is crucial here, as gradient descent is sensitive to feature magnitudes.
• Always preprocess your new input data the same way as training data.

---

✅ TODO

• Add support for mini-batch gradient descent.
• Add MSE/loss tracking over epochs.
• Save and load trained weights.
• Visualize convergence.

---

📚 License

MIT License. Free to use and modify.

---

Let me know if you'd like me to generate the file structure or write a setup.py or test case for this project.