README.md

metadata

language:
  - en
tags:
  - telecom
  - fraud
  - nigeria
  - cyber
size_categories:
  - 10K<n<100K

🇳🇬 Nigerian Telecom Fraud Analysis

Overview

This project presents an end-to-end Exploratory Data Analysis (EDA) of telecom-related fraud incidents in Nigeria. The goal is to uncover patterns, quantify financial impact, and provide actionable recommendations to mitigate fraud.

The analysis is based on a dataset containing ~50,000 fraud records, including fraud type, timestamps, operators, and financial impact.

---

Objectives

• Understand the distribution of fraud types
• Identify the most financially damaging fraud categories
• Compare fraud activity across telecom operators
• Provide business-oriented recommendations

---

Dataset Description

Key columns in the dataset:

• fraud_id – Unique identifier for each fraud case
• detected_at – Timestamp when fraud was detected
• fraud_type – Category of fraud (e.g., phishing, account takeover)
• operator – Telecom operator involved
• affected_phone – Victim phone number
• customer_id – Customer identifier
• financial_loss_ngn – Estimated financial loss in Nigerian Naira

---

Data Cleaning

Steps performed:

• Converted detected_at to datetime format
• Handled missing values in timestamps
• Standardized categorical values (fraud types, operators)
• Removed inconsistencies and duplicates where necessary

---

Exploratory Data Analysis

1. Fraud Type Distribution

• The most common fraud types include:

  • call forwarding fraud
  • identity theft

• other frauds occur in Relatively similar frequency.

2. Financial Impact Analysis

• Highest average losses:

  • Account Takeover
  • Subscription Fraud

• Moderate impact:

  • Phishing
  • Call forwarding fraud

• Low impact:

  • IMEI spoofing and similar technical frauds

Insight: Frequency does not necessarily equal severity. Some rare frauds are far more damaging.

3. Operator Comparison

• Fraud cases are distributed across multiple operators
• Some operators(MTN and Airtel) show higher exposure to specific fraud types

Insight: Targeted prevention strategies may be needed per operator.

---

3. Fraud location Comparison

• Fraud cases are distributed across multiple locations
• It is evident that most fraud cases occur in Lagos/

Insight: Lagos is seriously suspected of fraud.

Key Insights

• A small number of fraud types account for the majority of financial loss
• High-frequency fraud is not always high-impact
• Fraud patterns suggest both opportunistic and organized activity

---

Recommendations

Focus on High-Impact Fraud

• Prioritize detection systems for:

  • Account takeover
  • Subscription fraud

• Allocate more resources to prevent high-loss incidents

Improve Detection Systems

• Implement anomaly detection for unusual account behavior
• Use machine learning models for early fraud identification

Strengthen Customer Protection

• Introduce multi-factor authentication (MFA)
• Send real-time alerts for suspicious activities

Operator-Specific Strategies

• Customize fraud prevention measures per telecom operator
• Monitor operator-specific fraud trends

Continuous Monitoring

• Build dashboards for real-time fraud tracking
• Regularly update models and rules

---

Tools & Technologies

• Python (Pandas, NumPy)
• Data Visualization (Matplotlib, Seaborn)
• Jupyter Notebook

---

How to Use

1. Clone the repository
2. Open the notebook in Jupyter
3. Run all cells sequentially
4. Explore visualizations and insights

---

Future Work

• Build predictive models for fraud detection
• Incorporate real-time streaming data
• Add geolocation-based analysis
• Deploy as an interactive dashboard

---

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests.

---

License

This project is for educational and analytical purposes.

---

Data Source

The dataset used in this project was sourced from [nigerian-telecom-fraudulent-activity-datasets / huggingface].

Acknowledgments

This analysis was conducted as part of a Data Science assignment at Reichman University. The dataset is used for educational purposes only.