AMAZON RATINGS AND DISCOUNT STRATEGY

Personal project | Data analysis and modelling | 2026

Analysis of around 1,000 Amazon product listings to examine whether heavy discounting is associated with lower customer ratings, and whether a product's rating can be predicted from its price, discount and category. The business question came from a pricing and category management perspective: is discounting masking weak product quality, or are discounts and satisfaction largely independent?

Key contributions:

Cleaned the raw Kaggle dataset: dropped identifier and free-text columns, collapsed 221 raw category values into 9 top-level categories and coerced price, discount and rating fields to numeric types.

Explored the relationship between discount percentage and rating with correlation analysis, regression plots and category-level breakdowns.

Compared Linear Regression and Random Forest models predicting rating from price, discount and category, both meeting the RMSE target of 0.5.

Interpreted the results critically: low R squared values showed that price, discount and category are weak predictors of rating, so discounting and satisfaction appear largely independent.

Built a complementary Amazon sales report in Power BI, delivered as a PBIX file and PDF.

Collaborators: Gabriela Słomiany

Software Used: Jupyter Notebook, pandas, scikit-learn, kagglehub, Power BI

Programming Languages: Python, DAX

Repository: github.com/imgabrielas/Amazon-sales

Project Date: August 2026