shiragraiver/nigerian-retail-sample-30k-eda

---
language:
- en
size_categories:
- 10K<n<100K
---
# Nigerian Retail and E-Commerce Customer Segmentation – EDA Project

## Project Overview
This project presents an Exploratory Data Analysis (EDA) of the **Nigerian Retail and E-Commerce Customer Segmentation Data** dataset.
The goal of the project was to explore the structure of the data, identify important patterns, and examine whether customer total spending can be predicted using key numerical variables.

## Video Overview
<video src="https://huggingface.co/datasets/shiragraiver/nigerian-retail-sample-30k-eda/resolve/main/data_science_video.mp4" controls="controls" style="max-width: 720px;"></video>
## Dataset Description
This project uses the **Nigerian Retail and E-Commerce Customer Segmentation Data** dataset from **Hugging Face**.
The original dataset contains **150,000 rows and 11 features**, but for this project I used a random sample of **30,000 rows**.
The dataset includes numerical features such as `avg_order_value_ngn`, `total_orders`, `total_spend_ngn`, `last_purchase_days_ago`, and `lifetime_value_ngn`, as well as categorical features such as `segment`, `purchase_frequency`, `churn_risk`, `preferred_category`, and `seasonal_buyer`.

## Research Question
**Can customer total spending be predicted using numerical variables such as average order value, total number of orders, and days since the last purchase?**

## Target Variable
**`total_spend_ngn`**

## EDA Summary
The dataset was checked for duplicates, missing values, outliers, skewness, and relationships between variables.
No missing values were found in the sampled dataset.
Outliers were found mainly in `avg_order_value_ngn` and `total_spend_ngn`, and were treated using capping.
A log transformation was applied to the main spending-related variables in order to reduce skewness and improve interpretability.

## Research Questions and Answers

### 1. Is total spending related to average order value?
Yes. The analysis showed a clear positive relationship between `avg_order_value_ngn` and `total_spend_ngn`, suggesting that customers with higher average order values also tend to have higher total spending.

### 2. Is total spending related to the total number of orders?
The relationship appears relatively weak. High and low spending values appear across almost the entire range of order counts, suggesting that `total_orders` alone is not a strong predictor of total spending.

### 3. Is total spending related to days since the last purchase?
No clear relationship was found. Spending values remain widely scattered across the full range of `last_purchase_days_ago`, suggesting that purchase recency alone is not a strong predictor of total spending.

### 4. Which numerical variable is most strongly related to total spending?
`avg_order_value_ngn` was by far the strongest numerical variable related to `total_spend_ngn`, while the other numerical variables had very weak correlations.

## Key Insights
- `avg_order_value_ngn` is the most informative numerical variable for understanding customer total spending.
- `total_orders` and `last_purchase_days_ago` are much less informative on their own.
- Spending-related variables contained outliers and skewness, so capping and log transformation were important preprocessing decisions.
- The sampled dataset is complete and contains no missing values.

## Main Decisions
- Used a random sample of **30,000 rows**
- Kept all rows and treated outliers using **capping**
- Applied **log transformation** to skewed spending variables
- Focused the analysis mainly on numerical features relevant to predicting `total_spend_ngn`

## Limitations
- The dataset is **synthetic**, so it may not fully reflect real-world customer behavior.
- The analysis is exploratory and does not include a predictive model yet.

## Next Steps
A possible next step would be to build a **regression model** to test how accurately customer total spending can be predicted using the main numerical variables.

## Files Included
- `dataset_sample_30k.csv` – sampled dataset used in this project
- `eda_notebook.ipynb` – full EDA notebook
- `README.md` – project summary