Finlatics Sales Forecast Analysis

Project Overview

This project analyzes how advertising expenditures (TV, Radio, Newspaper) impact product sales using linear regression. The analysis answers key business questions and evaluates different modeling approaches.

Business Questions and Results

1. What is the average amount spent on TV advertising?

Result:
The average expenditure on TV advertising was $147.04
This indicates TV receives the largest share of the advertising budget.

2. What is the correlation between radio advertising and product sales?

Result:
The correlation between Radio advertising and Sales is 0.35
This moderate positive correlation suggests radio ads have some impact on sales.

3. Which advertising medium has the highest impact on sales?

Result:
TV advertising has the strongest correlation with sales (0.90)
This strong correlation indicates TV ads are the most effective medium for driving sales.

4. Linear Regression Model Performance

Results:

• R-squared: 0.906
• Mean Squared Error: 2.90
• Root Mean Squared Error: 1.70

Key Coefficients:

• TV: 0.0545 (most significant)
• Radio: 0.1010
• Newspaper: -0.0043

The model explains 89.9% of sales variance, with TV being the most influential predictor.

Visualization shows strong alignment between predicted and actual sales

5. Sales Prediction for New Expenditures

Input:

• TV: $200
• Radio: $40
• Newspaper: $50

Predicted Sales: 19.87 units
This helps budget allocation decisions for optimal sales impact.

6. Effect of Normalization

Result:
Normalization produced identical performance metrics:

• R-squared: 0.906
• MSE: 2.90

Normalization doesn't affect linear regression performance but aids interpretation when features have different scales.

7. Impact of Using Only Radio & Newspaper

Results:

• R-squared: 0.1097
• MSE: 27.51

Dramatic performance drop confirms TV's critical role in sales prediction.

Key Insights

1. TV advertising drives the most sales
2. Newspaper ads show negligible impact (negative coefficient)
3. The full model explains 90% of sales variation
4. Removing TV from predictors makes the model ineffective

How to Run

1. Clone the repository
2. Install requirements: pip install -r requirements.txt
3. Run the notebook: jupyter notebook sales_prediction.ipynb

Dependencies

• Python 3.7+
• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn

Files

• sales_prediction.ipynb: Jupyter notebook with full analysis
• advertising_sales_data.xlsx: Dataset