Sales Prediction using Python

CodeAlpha Data Science Internship — Task 4

Intern: Omokhoa Oshose Tosayoname
Intern ID: CA/DF1/71570
Duration: 20th May 2026 – 20th June 2026

---

Overview

This project predicts product sales based on advertising spend across three media channels: TV, Radio, and Newspaper. Using the classic Advertising dataset (200 observations), we explore how budget allocation across channels influences sales outcomes and build multiple regression models to forecast future sales.

Business Question: How does advertising spend across TV, Radio, and Newspaper channels drive sales, and which channel delivers the highest return?

---

Project Pipeline

Data Loading --> EDA & Visualisation --> Feature Engineering
    --> Model Training --> Evaluation --> Business Insights

---

Project Structure

CodeAlpha_SalesPrediction/
├── data/
│   └── Advertising.csv          # Raw dataset
├── notebooks/
│   └── sales_prediction.ipynb   # Main notebook (fully executed)
├── requirements.txt
└── README.md

---

Dataset

Feature	Description
TV	Advertising budget spent on TV (in $000s)
Radio	Advertising budget spent on Radio (in $000s)
Newspaper	Advertising budget spent on Newspaper (in $000s)
Sales	Units sold (in thousands) — target variable

---

Models Trained and Compared

Model	Features Used
Linear Regression	Base (TV, Radio, Newspaper)
Ridge Regression	Base
Lasso Regression	Base
Polynomial Regression (degree 2)	Base
Random Forest Regressor	Base + engineered features
XGBoost Regressor	Base + engineered features

Engineered features include: TV×Radio interaction, TV×Newspaper, Radio×Newspaper, Total Budget, TV Share, Radio Share.

---

Results

Model	R²	RMSE	MAE
Random Forest	0.9880	0.6148	0.4797
Polynomial Reg (d=2)	0.9869	0.6426	0.5262
XGBoost	0.9846	0.6980	0.5449
Linear Regression	0.8994	1.7816	1.4608
Ridge Regression	0.8988	1.7872	1.4643
Lasso Regression	0.8983	1.7913	1.4613

Best model: Random Forest (R² = 0.9880)

---

Key Business Findings

• TV advertising has the strongest correlation with sales (r ~ 0.78).
• Radio is the second most impactful channel; its interaction with TV is highly predictive.
• Newspaper spending shows the weakest relationship with sales outcomes.
• Companies in the highest TV budget quartile achieve nearly 3x the sales of those in the lowest quartile.
• Recommendation: Prioritise TV and Radio spend for maximum sales lift; reconsider Newspaper allocation.

---

Visualisations Produced

• Advertising budget distributions per channel
• Sales distribution and Q-Q plot
• Scatter plots: each channel vs sales with regression lines and correlation coefficients
• Correlation heatmaps (base and engineered features)
• Budget allocation pie chart and bar chart
• Box plots for all variables
• Pairplot of all features
• Engineered feature correlation matrix
• Model performance comparison (R², RMSE, MAE)
• Actual vs Predicted plots for top two models
• Residual analysis plots
• Random Forest feature importances
• OLS regression summary (statsmodels)
• Linear regression coefficients chart
• Sales segmentation by TV budget quartile
• Advertising channel ROI proxy

---

How to Run

1. Clone this repository:

   git clone https://github.com/Tosa9/CodeAlpha_SalesPrediction.git
   cd CodeAlpha_SalesPrediction

2. Install dependencies:

   pip install -r requirements.txt

3. Launch the notebook:

   jupyter notebook notebooks/sales_prediction.ipynb

---

Dataset Source

Advertising Dataset — Kaggle

---

CodeAlpha Data Science Internship | Task 4
#CodeAlpha #DataScience #MachineLearning #SalesPrediction #Python