shiraBASH/online-shoppers-eda

Exploratory Data Analysis (EDA) on the Online Shoppers Purchasing Intention Dataset

Author: Shira Bash

Project Overview

This project performs Exploratory Data Analysis (EDA) on the Online Shoppers Purchasing Intention dataset. The goal is to understand which behavioral patterns influence the likelihood that a website visitor completes a purchase (Revenue = True).

The analysis includes:

• —Data exploration & validation
• —Visualizations (histograms, scatter plots, box plot)
• —Statistical insights
• —Research questions & conclusions

Dataset Summary

The dataset contains 12,330 Rows and 18 Columns, all information about 12,000+ user sessions on an e-commerce website, including:

Numerical Features

• —Administrative_Duration – time spent on admin-related pages
• —ProductRelated_Duration – time spent viewing products
• —BounceRates – probability that a page causes the visitor to leave immediately
• —ExitRates – probability that a page is the last in the session
• —PageValues – estimated value of each page in the conversion process

Categorical Features

• —VisitorType (Returning / New / Other)
• —Month
• —Weekend (True/False)

Target Variable

• —Revenue – whether the session resulted in a purchase

Step 1: Data Cleaning

• —Checked for missing values (none found).
• —Verified numeric columns are in valid ranges (BounceRates/ExitRates between 0–1).
• —Ensured categorical labels are consistent.
• —Removed duplicates if needed.

Step 2: Key Visualizations

Histograms – Distribution of Key Features

Used to see behavior patterns (short vs. long visits, valued pages, etc.). → All distributions are right-skewed, meaning most users browse quickly, but a few take much longer.

• —Administrative_Duration – time spent on admin-related pages
• —ProductRelated_Duration – time spent viewing products
• —BounceRates – probability that a page causes the visitor to leave immediately
• —PageValues – estimated value of each page in the conversion process

Scatter Plot – BounceRates vs ExitRates

Shows the relationship between "leaving immediately" vs. "ending a session". Insight: Higher BounceRates strongly correlate with higher ExitRates. Purchases (Revenue=True) cluster at lower bounce and exit rates.

Box Plot – Looking for Outliers

Used to visualize extreme values in durations and PageValues. Outliers were kept because they represent genuine long browsing sessions that matter for conversions.

Step 3: Research Questions & Insights

Q1: Does visitor type affect purchase likelihood?

✔ Yes. new visitors have a much higher purchase rate than returning visitors. perhaps from targeted campaigns or specific searches.

Q2: Do purchases vary by month?

✔ Yes. November shows the highest purchase rate — consistent with Black Friday/Cyber Monday.

Q3: Does weekend activity impact purchase rates?

✔ Yes. Weekend sessions show a slightly higher purchase rate compared to weekday sessions, suggesting that users browsing during weekends are more likely to complete a purchase. This might be due to having more free time to explore or complete shopping decisions.

Key Insights Summary

• —Most traffic is fast/short, but buyers tend to explore longer.
• —New visitors tend to purchase more than others .
• —PageValues is one of the strongest indicators of conversion.
• —Lower BounceRates = higher purchase probability.
• —Seasonality matters (November peak).

Files Included

• —`online_shoppers_intention.csv` - dataset used for the analysis
• —assignment_shira_bash.ipynb – notebook containing full Python EDA process
• —README.md – summary of results, questions, and insights
• —`Loom Video` - project walkthrough presentation