Sales Prediction using Simple Linear Regression

Enhanced, Modular Implementation with Object-Oriented Design

Problem Statement

Build a model that predicts sales based on the amount of money spent on different platforms for marketing. This implementation focuses specifically on analyzing the relationship between TV advertising spend and Sales using a simple linear regression model.

The dataset used is the Advertising dataset from the Introduction to Statistical Learning (ISLR) book.

---

Internship Context

This project is my second task for the CodSoft Data Science Internship. I have already completed Task 1 – Titanic Survival Prediction – and will be completing two more tasks to fulfill the internship requirements.

---

About This Implementation

The original script provided with the dataset was a basic, linear version written in notebook format. I took that as a base and developed a modular, reusable, and production-ready version using Python and the scikit-learn library.

The logic has been restructured using object-oriented programming, where all key steps—data loading, model training, performance evaluation, visualization, and prediction—are handled through a well-organized class.

---

Key Features

Object-Oriented Design

Encapsulates the entire analysis into a single class LinearRegressionAnalysis, allowing for:

• Reusability across datasets
• Cleaner separation of concerns
• Easier debugging and extension

Flexible Data Input

The implementation supports any CSV dataset with numeric input and target columns. Users only need to specify:

x_column = 'TV'
y_column = 'Sales'

Error Handling

Includes try-except blocks for:

• CSV file reading
• Column validation

This ensures the program provides helpful messages and doesn't crash due to bad input.

Performance Evaluation

Evaluates the model using:

• R² Score (training and testing)
• RMSE
• MAE Includes interpretation logic to explain results.

Visualization

Generates four visualizations in one figure:

1. Actual vs Predicted values
2. Regression line on training data
3. Residuals vs Predicted values
4. Distribution of residuals

Easy Prediction

Accepts new input values and returns predicted outputs:

model.make_predictions([100, 150, 200])

---

Example Usage

from regression_analysis import LinearRegressionAnalysis

model = LinearRegressionAnalysis()
model.load_data('advertising.csv')
model.train_model(x_column='TV', y_column='Sales')
model.evaluate_performance()
model.plot_results()
model.make_predictions([100, 150, 200])

---

Requirements

Install the required libraries with:

pip install pandas scikit-learn matplotlib seaborn

---

Dataset

This project uses the Advertising dataset, typically containing:

• TV advertising spend
• Radio advertising spend
• Newspaper advertising spend
• Product sales

Only the TV and Sales columns are used for this linear regression model.

---

Acknowledgment

To understand the core concepts of linear regression and improve this project, I referred to:

• YouTube tutorials for practical implementations
• The Data Science Masterclass by Krish Naik, which helped clarify the statistical and modeling aspects

These resources provided valuable guidance, especially in structuring the model evaluation and visualization parts.