electricsheepafrica/nigerian-banking-ussd-banking-logs

USSD Banking Logs (*737#, *901#, etc.)

Dataset Description

Nigerian USSD banking (*99#) logs with fraud labels

This is a production-grade synthetic dataset with authentic Nigerian banking context, designed for USSD fraud detection, feature phone banking optimization, and accessibility analytics.

Key Features

• 🇳🇬 Nigerian Context: Real bank names, states, demographics
• 📊 ML-Ready Labels: Fraud flag (1.2% prevalence)
• ⚡ Large Scale: 2,000,000 rows
• 🎯 Realistic Distributions: Short session durations, menu depth distributions, transaction type patterns
• 📅 Temporal Coverage: 2023-01-01 to 2024-12-31

Dataset Structure

Splits

• Train: 1,400,000 rows (70%)
• Validation: 300,000 rows (15%)
• Test: 300,000 rows (15%)

Schema

Key columns: session_id, timestamp, customer_id, phone_number, ussd_code, menu_depth, session_duration_seconds, transaction_type, session_status, fraud_flag

Usage

from datasets import load_dataset

# Load full dataset
dataset = load_dataset("electricsheepafrica/nigerian_ussd_banking_logs")

# Load specific split
train_data = load_dataset("electricsheepafrica/nigerian_ussd_banking_logs", split="train")

Labels

• Type: Fraud flag (1.2% prevalence)
• Use Case: USSD fraud detection, feature phone banking optimization, and accessibility analytics

Nigerian Context

Nigerian bank USSD codes (*737# GTBank, *901# Access, *894# First Bank, etc.), typical menu navigation patterns, timeout behaviors, and rural/informal user behaviors.

Citation

@dataset{nigerian_banking_ussd_banking_logs,
  title = {Nigerian USSD Banking Logs (*737#, *901#, etc.)},
  author = {Electric Sheep Africa},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/electricsheepafrica/nigerian_ussd_banking_logs}}
}

License

Apache 2.0 - Free for commercial and research use.

Contact

• Organization: Electric Sheep Africa
• Collection: Nigeria Banking & Finance Datasets