linear-regression-from-scratch

**Type of Work:**

A desktop application for training a Linear Regression model using Python, built entirely from scratch without relying on scikit-learn's built-in training functions.

**Features:**

- Implementation of both Batch Gradient Descent and Stochastic Gradient Descent from scratch

- Manual train/test splitting without using sklearn's built-in split function

- Feature standardization and L2 regularization to reduce overfitting

- Interactive GUI (Tkinter) displaying a real-time training loss curve

- Performance evaluation using MSE, RMSE, and R² on both training and test sets

**Implementation Approach:**

The algorithm was built using only NumPy, with gradients computed manually and weights updated iteratively across epochs. The training logic was separated from the user interface, so the GUI simply calls the underlying mathematical functions and displays results interactively to the user in real time during training.