electricsheepafrica/Nigerian-Financial-Transactions-and-Fraud-Detection-Dataset

Nigerian Financial Fraud Detection Dataset (Enhanced)

Overview

This is a comprehensive synthetic financial fraud detection dataset specifically engineered for the Nigerian fintech ecosystem. The dataset contains 5,000,000 transactions with 45 advanced features including sophisticated user behaviour analytics, device intelligence, risk scoring, and temporal patterns tailored for Nigerian financial fraud detection.

We have found that a lot of people are unable to use the full dataset because of compute constraints. If you are splitting the dataset, make sure you specify the ratio of fraud to non-fraud signals within the amount that you collect. The sweet spot is 27%. If you do not do this, your script might select unintelligently, and you might not get enough fraud signals to get a great model.

Key Highlights

• 🇳🇬 Nigerian-Localized: Currency (NGN), cities, payment channels, IP ranges
• 🧠 Advanced Features: 30 sophisticated fraud detection features
• 📊 Production-Ready: No data leakage, optimized for ML training
• 🔍 Behavioral Analytics: User personas, transaction patterns, device intelligence
• ⚡ High-Performance: Engineered for 5M+ transaction processing

Dataset Statistics

• Total Transactions: 5,000,000
• Total Features: 45
• Fraud Rate: ~15% (realistic for emerging markets)
• Time Span: Simulated 12-month period
• Unique Users: ~500,000 sender accounts
• Nigerian Cities: 20 major cities across 6 geo-regions

Feature Categories

🏦 Core Transaction Features (15 features)

• transaction_id: Unique transaction identifier
• sender_account: Anonymized 10-digit Nigerian account number
• receiver_account: Anonymized 10-digit Nigerian account number
• amount_ngn: Transaction amount in Nigerian Naira (NGN)
• timestamp: Transaction timestamp with Nigerian business patterns
• payment_channel: Nigerian payment methods (USSD, Mobile App, Card, Bank Transfer)
• merchant_category: Localized merchants (Jumia, MTN Airtime, Bet9ja, etc.)
• location: Major Nigerian cities
• ip_address: IP addresses from Nigerian IP ranges
• device_hash: Anonymized device identifier
• is_fraud: Binary fraud label (0=legitimate, 1=fraud)
• fraud_type: Specific fraud categories (Account Takeover, Identity Fraud, etc.)
• bvn_linked: Bank Verification Number linkage status
• sender_persona: Behavioral profile (Salary Earner, Student, Trader)
• geospatial_velocity_anomaly: Impossible travel pattern flag

👤 User Behavior Features (5 features)

• user_avg_txn_amt: Average transaction amount per user (expanding window)
• user_std_txn_amt: Standard deviation of user's transaction amounts
• user_top_category: Most frequent merchant category per user
• user_txn_frequency_24h: Transaction frequency indicator
• user_txn_count_total: Lifetime transaction count per user

📱 Device & IP Intelligence (5 features)

• device_seen_count: Total transactions from each device
• is_device_shared: Flag for devices used by multiple users (0/1)
• ip_seen_count: Total transactions from each IP address
• is_ip_shared: Flag for IPs used by multiple users (0/1)
• ip_geo_region: Nigerian geo-region (South West, North Central, etc.)

📈 Transaction History/Window Features (4 features)

• txn_count_last_1h: Recent transaction count (expanding window)
• txn_count_last_24h: 24-hour transaction count (expanding window)
• total_amount_last_1h: Sum of recent transactions (expanding window)
• avg_gap_between_txns: Average time between user transactions (minutes)

⚠️ Risk Scoring Fields (4 features)

• merchant_fraud_rate: Historical fraud rate per merchant category (no leakage)
• channel_risk_score: Risk score by payment channel (USSD=0.8, Mobile=0.6, etc.)
• persona_fraud_risk: Risk by user persona (Trader=0.7, Student=0.5, etc.)
• location_fraud_risk: Historical fraud rate per Nigerian city (no leakage)

🕐 Temporal Features (4 features)

• txn_hour: Transaction hour (0-23)
• is_weekend: Weekend transaction flag (0/1)
• is_salary_week: Last 5 days of month flag (0/1)
• is_night_txn: Night transaction flag 11pm-5am (0/1)

🔧 Technical Features (8 features)

• new_device_for_sender: First-time device usage flag
• shared_device_hash: Device used by multiple accounts
• time_since_last: Minutes since user's last transaction
• Various derived flags and indicators

Nigerian Localization Details

🏙️ Geographic Coverage

6 Nigerian Geo-Regions Represented:

• South West: Lagos, Ibadan, Abeokuta, Oyo, Ogbomoso
• North Central: Abuja, Jos, Ilorin, Okene
• North West: Kano, Zaria, Kaduna, Sokoto
• South South: Port Harcourt, Benin City, Warri
• South East: Aba, Enugu, Onitsha
• North East: Maiduguri

💳 Payment Channels (Risk-Weighted)

• USSD (Risk: 0.8) - Most common, highest fraud risk
• Mobile App (Risk: 0.6) - Moderate risk
• Card (Risk: 0.4) - Lower risk
• Bank Transfer (Risk: 0.3) - Lowest risk

🛒 Merchant Categories (Nigerian Context)

• Jumia Purchase, Konga Shopping, MTN Airtime Top-up
• Airtel Data Bundle, Bet9ja Stake, NairaBet Gaming
• Uber Ride, Bolt Transport, Flutterwave Payment
• Paystack Transaction, Opay Transfer, PalmPay Service
• And 10+ other Nigerian-specific merchants

👥 User Personas (Behavioral Profiles)

• Salary Earner (Risk: 0.4): Regular monthly patterns, moderate amounts
• Student (Risk: 0.5): Small frequent transactions, education-related
• Trader (Risk: 0.7): High-volume, irregular patterns, higher risk

Fraud Types and Scenarios

🚨 Fraud Categories Included

1. Account Takeover: Compromised accounts with unusual patterns
2. Identity Fraud: Fake accounts, unlinked BVN
3. Impossible Travel Fraud: Geospatial velocity anomalies
4. SIM Swap Fraud: Device changes with suspicious activity
5. Card-Not-Present: Online fraud without physical card
6. Deposit Fraud: Fake deposit schemes
7. Money Laundering: Structured transactions, unusual patterns

🎯 Fraud Detection Signals

• BVN linkage status (unlinked = higher risk)
• New device usage patterns
• Shared device/IP indicators
• Geospatial velocity anomalies
• Off-hours transaction patterns
• Unusual merchant category combinations
• Rapid transaction sequences

Technical Implementation

🔒 Data Privacy & Security

• All account numbers are randomly generated (no real accounts)
• Device hashes are anonymized identifiers
• IP addresses are from public Nigerian ranges
• No personally identifiable information (PII) included

📊 Feature Engineering Methodology

• No Data Leakage: Risk scores use expanding windows with .shift(1)
• Temporal Consistency: Features computed chronologically
• Performance Optimized: Efficient groupby operations for 5M+ rows
• Realistic Patterns: Based on Nigerian fintech transaction behaviors

🔄 Rolling Window Logic

# Example: Merchant fraud rate without leakage
merchant_fraud_rate = df.groupby('merchant_category')['is_fraud']
                       .transform(lambda x: x.expanding().mean().shift(1).fillna(0.1))

Use Cases & Applications

🎯 Primary Use Cases

1. Fraud Detection Model Training: Supervised learning with rich feature set
2. Risk Scoring System Development: Real-time transaction scoring
3. Behavioral Analytics: User pattern analysis and segmentation
4. Anomaly Detection: Unsupervised fraud detection research
5. Nigerian Fintech Research: Emerging market fraud patterns

🏢 Target Industries

• Nigerian fintech companies (Flutterwave, Paystack, Opay, etc.)
• Traditional banks expanding digital services
• Fraud detection technology vendors
• Academic research institutions
• Regulatory bodies and compliance teams

📈 Model Performance Expectations

• Baseline Accuracy: 85-90% with basic features
• Enhanced Accuracy: 92-95% with advanced features
• Precision/Recall: Optimized for Nigerian fraud patterns
• Real-time Scoring: Features designed for low-latency inference

Data Quality & Validation

✅ Quality Assurance

• Completeness: No missing values in critical features
• Consistency: Logical relationships maintained across features
• Realism: Patterns based on Nigerian financial behaviors
• Balance: Appropriate fraud/legitimate transaction ratio

🔍 Validation Metrics

• Transaction amounts follow realistic Nigerian distributions
• Temporal patterns align with Nigerian business hours
• Geographic distribution matches Nigerian population centers
• Fraud patterns consistent with emerging market trends

Limitations & Considerations

⚠️ Important Limitations

1. Synthetic Data: Not real transactions, patterns may differ from reality
2. Temporal Scope: Limited to 12-month simulation period
3. Feature Completeness: May not capture all real-world fraud signals
4. Regional Focus: Specific to Nigerian context, may not generalize
5. Fraud Evolution: Real fraud patterns evolve faster than synthetic data

🎯 Recommended Usage

• Use as training data supplement, not replacement for real data
• Validate model performance on real Nigerian transaction data
• Regular model retraining as fraud patterns evolve
• Combine with external data sources for production systems

Technical Specifications

📋 File Details

• Format: CSV (Comma-separated values)
• Size: ~2.5GB uncompressed
• Encoding: UTF-8
• Delimiter: Comma (,)
• Header: Yes (first row contains column names)

🔧 System Requirements

• Memory: 8GB+ RAM recommended for full dataset loading
• Storage: 5GB+ available space
• Processing: Multi-core CPU recommended for feature engineering
• Software: Python 3.7+, pandas 1.3+, numpy 1.20+

Getting Started

📚 Quick Start Example

import pandas as pd
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# Load the dataset
df = pd.read_csv('financial_fraud_detection_dataset_nigeria.csv')

# Basic exploration
print(f"Dataset shape: {df.shape}")
print(f"Fraud rate: {df['is_fraud'].mean():.2%}")
print(f"Features: {df.columns.tolist()}")

# Feature selection for modeling
feature_cols = [col for col in df.columns if col not in ['transaction_id', 'is_fraud', 'fraud_type']]
X = df[feature_cols]
y = df['is_fraud']

# Train/test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Simple model training
model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)

# Evaluate
accuracy = model.score(X_test, y_test)
print(f"Model accuracy: {accuracy:.2%}")

🎯 Advanced Analytics Examples

# Analyze fraud patterns by region
region_fraud = df.groupby('ip_geo_region')['is_fraud'].agg(['count', 'mean'])
print("Fraud rates by Nigerian region:")
print(region_fraud)

# User behavior analysis
user_stats = df.groupby('sender_persona').agg({
    'amount_ngn': ['mean', 'std'],
    'is_fraud': 'mean',
    'user_txn_count_total': 'mean'
})
print("\nUser persona analysis:")
print(user_stats)

# Temporal fraud patterns
temporal_fraud = df.groupby(['txn_hour', 'is_weekend'])['is_fraud'].mean()
print("\nTemporal fraud patterns:")
print(temporal_fraud)

Citation & Attribution

📄 Recommended Citation

Nigerian Financial Fraud Detection Dataset (Enhanced)
Synthetic dataset for fraud detection research in Nigerian fintech
Generated: 2024
Features: 45 advanced fraud detection features
Transactions: 5,000,000 synthetic transactions

🏷️ Tags

fraud-detection nigeria fintech machine-learning synthetic-data behavioral-analytics risk-scoring emerging-markets financial-crime anomaly-detection

Support & Updates

📞 Contact Information

For questions, issues, or collaboration opportunities regarding this dataset, please refer to the project documentation or contact the development team.

🔄 Version History

• v2.0 (Current): Enhanced with 30 advanced features, Nigerian localization
• v1.0: Basic fraud detection dataset with core features

🚀 Future Enhancements

• Transaction network analysis features
• Additional Nigerian payment channels
• Seasonal fraud pattern variations
• Enhanced merchant category coverage
• Real-time streaming data simulation