ajay-anil-kumar/Globo-Mart

GloboMart: A Synthetic Retail Analytics Dataset

Dataset Summary

GloboMart is a comprehensive, synthetic dataset simulating five years of operations for a fictional global retailer. It's designed to provide a realistic and analytically deep resource for data analysts, scientists, and engineers. 📊

The dataset captures the entire customer journey, from marketing interactions and web engagement to sales transactions and final fulfillment. This end-to-end scope enables a wide range of analyses to answer critical business questions, such as:

• Marketing: How effective are our campaigns? (Dim_Marketing, Fact_Web_Analytics)
• Web Engagement: What is the user behavior on our website? (Fact_Web_Analytics)
• Profitability: What are the main drivers of sales and profit? (Fact_Sales)
• Logistics: How can we improve supply chain efficiency? (Dim_Shipments)

This dataset is ideal for practicing SQL, developing BI dashboards, and building predictive models in a realistic retail context.

---

Schema and Data Structure

Star Schema Architecture

The data warehouse is structured as a classic star schema, which is optimized for high-performance analytical queries. This design features central Fact tables (containing quantitative transactional data) that are connected to surrounding Dimension tables (containing descriptive, contextual attributes).

File Structure and Tables

The dataset is provided in the Apache Parquet format for efficient storage and fast read performance. It consists of the following tables:

• Dimension Tables (Stored as single Parquet files):

  • dim_customers.parquet: Detailed information about each customer.
  • dim_date.parquet: A date dimension to facilitate time-based analysis.
  • dim_marketing.parquet: Data on marketing campaigns.
  • dim_products.parquet: Attributes for each product, such as category.
  • dim_shipments.parquet: Information related to order fulfillment and shipping.
  • dim_stores.parquet: Details about each retail store location.

• Fact Tables (Partitioned by year and month for scalability):

  • fact_sales/: Contains all individual sales transaction records.
  • fact_web_analytics/: Records of user sessions and events on the website.

The partitioned structure of the fact tables (/year=YYYY/month=MM/) allows for efficient querying of specific time periods, as data loaders can prune partitions without reading the entire dataset.

---

How to Use the Dataset

You can easily load the tables using the 🤗 datasets library. First, ensure you have the necessary libraries installed:

pip install datasets pyarrow

Loading a Dimension Table

Dimension tables are single files and can be loaded directly.

from datasets import load_dataset

# The Hugging Face repository ID
repo_id = "ajay-anil-kumar/Globo-Mart"

# Load the products dimension table
products_ds = load_dataset(repo_id, data_files="globomart_dataset/dim_products.parquet")

print("Products Table Schema:")
print(products_ds['train'].features)
print("\nFirst Product Record:")
print(products_ds['train'][0])

Loading a Partitioned Fact Table

The datasets library can automatically load all partitions of a fact table by using a glob pattern.

from datasets import load_dataset

repo_id = "ajay-anil-kumar/Globo-Mart"

# Load the entire 'fact_sales' table by pointing to all .parquet files within the directory
sales_ds = load_dataset(repo_id, data_files="globomart_dataset/fact_sales/**/*.parquet")

print(f"Total sales records: {sales_ds['train'].num_rows}")
print("\nFirst Sales Record:")
print(sales_ds['train'][0])

Dataset Creation

The dataset was programmatically generated using a Python-based simulation engine to ensure temporal consistency and analytical realism. The simulation ran on a day-by-day basis over a five-year period, incorporating realistic business patterns:

• Seasonality: Sales volumes for relevant product categories (e.g., ‘Electronics’) were increased in corresponding quarters (e.g., Q4).

• Event-Driven Spikes: Sales were multiplied on weekends and holidays to simulate peak shopping periods.

• Customer & Product Popularity: A power-law distribution was used to ensure a realistic concentration of sales among a small number of popular customers and products.

• Correlated Events: Website conversion events in Fact_Web_Analytics are directly linked to transactional records in Fact_Sales, creating a coherent user journey.

Additional Information

Licensing

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation

If you use this dataset in your work, please consider citing it:

@misc{ajay_anil_kumar_2025,
    author       = { Ajay Anil Kumar and Rohit Kumar },
    title        = { Globo-Mart (Revision d24f020) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/ajay-anil-kumar/Globo-Mart },
    doi          = { 10.57967/hf/6676 },
    publisher    = { Hugging Face }
}

Dataset Curators

This dataset was created by Ajay Anil Kumar and Rohit Kumar. For questions or contributions, please open an issue in the repository.