04 / DATA ANALYSIS / E-COMMERCE
Amazon Sales Analysis
An end-to-end data analysis project on Amazon India sales, combining data cleaning, exploratory visualization, order-cancellation modeling, and an interactive Streamlit presentation.
The dataset
The project uses the Amazon Sales Report dataset from Kaggle, containing sales information from Amazon India between April and June 2022 .
The goal was broader than creating charts: first make the dataset analysis-ready, then explore commercial patterns, and finally investigate whether an order rejection/cancellation event could be predicted.
Data preparation
- Removed columns with little analytical value or effectively constant information.
- Handled missing values in fields such as courier status, promotion identifiers, amount, and shipping information.
- Renamed and transformed variables to make the dataset easier to analyze, including customer type and order amount.
- Created a month variable for aggregation and ordered product sizes for more meaningful analysis.
Analysis and modeling
Explore commercial behavior
Build visual analyses around net revenue, average monthly order amount, top products, sales by product size, sales over time, and category-level order trends.
Predict rejected orders
Frame order rejection as a classification problem and compare Logistic Regression with Random Forest.
Understand the drivers
Use the classification workflow not only to predict the event, but also to investigate which features contribute most to the outcome.
Present the work interactively
Package the analysis into a Streamlit presentation so users can explore results without opening notebooks directly.
The project combines three skills that are often separated in coursework: preparing messy business data, extracting readable insights, and building a predictive model around a concrete event.