A research-style EDA spanning three e-commerce datasets — eBay (UK orders + product listings), Flipkart (India), and a Pakistan-wide e-commerce dataset. Rather than a single-dataset exploration, this project asks comparative questions across platforms and markets: how much do they sell, at what prices, how are customers rating products, and does festival season actually drive sales?
Can data from real e-commerce platforms prove the business case for digital commerce? The notebook frames itself around a central thesis: digital marketing and e-commerce dramatically amplify a company’s sales reach — and sets out to support it with concrete statistics on volume, geography, pricing, and customer sentiment across multiple platforms and countries.
Four datasets were loaded and independently cleaned — a substantial preprocessing effort given messy price strings (currency symbols, commas), inconsistent date formats, and string-encoded numeric fields. Key analytical threads included:
Geographic reach: Drilling from country → county → city to find where eBay orders concentrated (spoiler: London dominated at every level, including a data-cleaning catch where “london” and “London” were being counted separately).
Pricing: Raw means were wildly skewed by outliers in both eBay and Flipkart datasets. A consistent IQR-based trimming strategy (cutting above the 75th percentile) was applied to surface more representative average prices — ~$98 for eBay, ~913 INR for Flipkart, ~845 PKR for Pakistan. Flipkart’s average discount rate came out at ~52%.
Seasonality vs. festivals: Pakistan’s 2017 sales data was combined with a scraped public holiday calendar (fetched live via pd.read_html) to test whether festival days drive above-average sales. The answer was mostly no — only two of the five peak sales months overlapped with the five most festival-heavy months.
Customer sentiment: Feedback and ratings were bucketed into positive/neutral/negative across eBay (order feedback) and both eBay and Flipkart (product ratings). eBay came out at ~95.6% positive; Flipkart’s numbers were lower but described as “more realistic.”
Brand vs. sales: Top-rated brands on eBay (Apple, Samsung, Sony) aligned closely with top-selling brands. On Flipkart, the two best-selling brands (Allure Auto, Voylla) had no ratings data and below-average ratings respectively — suggesting brand loyalty plays out differently across markets.
Volume benchmarks: Daily sales rates were computed for all three platforms — ~156/day for eBay, ~157/day for Flipkart, ~741/day across Pakistan’s market. November consistently emerged as the top month across both Pakistan years in the dataset.
Outliers in price data are the rule, not the exception, in real e-commerce datasets — and the choice of how to handle them (trim vs. transform vs. cap) meaningfully changes the headline number. Scraping live reference data mid-notebook (the Pakistani holiday calendar) to cross-validate an assumption was a useful pattern: rather than guessing whether festivals drive sales, the data was brought in to test it directly. The cross-platform structure also surfaced how easy it is to over-generalize from one market — the brand-loyalty finding on eBay simply didn’t hold on Flipkart.