Telecom Fraud Detection Dataset

Each record represents an outgoing phone call along with historical caller behavior, network reputation indicators, and fraud-related signals commonly used by telecom operators and cybersecurity systems. The dataset has been carefully designed to simulate realistic fraud patterns while maintaining overlapping characteristics between legitimate and fraudulent calls, making it suitable for robust machine learning experimentation.

Unlike simplified datasets where fraud can be identified using a few obvious rules, this dataset includes challenging edge cases such as high-volume legitimate callers and low-activity fraudulent numbers. This encourages models to learn meaningful relationships between features rather than relying on simple thresholds.

Dataset Overview

The dataset consists of 10,000 call records with 15 columns, including 14 predictive features and 1 target variable.

Dataset Highlights

• Total Records: 10,000
• Features: 14
• Target Variable: fraud_label
• Class Distribution: Approximately 70% Legitimate and 30% Fraudulent
• Missing Values: None
• Duplicate Records: None
• File Format: CSV
• Dataset Type: Fully Synthetic
• Machine Learning Ready: Yes

Dataset Features

The dataset combines multiple categories of telecom intelligence to provide a comprehensive view of caller behavior.

Caller Behavior

Behavioral features describe how each phone number interacts with the network over time.

These include:

• Caller Age (Days)
• Calls Per Day
• Call Duration
• Average Historical Call Duration
• Night Call Ratio

These variables help identify unusual calling patterns that may indicate suspicious activity.

Reputation Indicators

Reputation-based metrics capture how trustworthy a phone number appears based on historical interactions.

Available features include:

• Receiver Block Rate
• Spam Reports Count
• Reputation Score
• Country Code Risk Score

These indicators allow models to distinguish between trusted callers and potentially fraudulent numbers.

Graph Intelligence Features

Modern fraud detection systems often analyze relationships between callers rather than evaluating each call independently.

Graph-based features include:

• Unique Receivers (24 Hours)
• Graph Degree
• Previous Fraud Associations
• Sequential Dialing Score

These variables provide additional context about caller connectivity and communication behavior.

Why This Dataset Is Valuable

The Telecom Fraud Detection Dataset has been created to resemble real-world fraud detection challenges rather than simplified academic examples.

Some of its key characteristics include:

• Realistic statistical feature distributions
• Overlapping fraud and legitimate behavior
• No perfectly separable classes
• Correlated behavioral and reputation features
• High-volume legitimate callers
• Low-volume stealth fraud cases
• Machine learning-friendly feature design

Potential Machine Learning Applications

The dataset can be used for a wide variety of AI and data science projects, including:

Fraud Detection

Develop binary classification models capable of identifying fraudulent telecom calls.

Feature Engineering

Create new behavioral and reputation-based features to improve model performance.

Class Imbalance Learning

Experiment with techniques such as:

• SMOTE
• Class Weighting
• Threshold Optimization
• Sampling Strategies

Model Explainability

Analyze prediction behavior using:

• SHAP Values
• Feature Importance
• Permutation Importance

Exploratory Data Analysis

Explore:

• Feature Distributions
• Correlation Analysis
• Outlier Detection
• Fraud Pattern Discovery

Ideal For

This dataset is suitable for:

• Data Scientists
• Machine Learning Engineers
• Cybersecurity Researchers
• Telecom Analytics Teams
• Students & Educators
• AI Portfolio Projects
• Fraud Detection Research
• Business Intelligence Applications

Data Quality

The dataset has been generated with machine learning applications in mind and requires minimal preprocessing.

Quality characteristics include:

• No missing values
• No duplicate records
• Consistent feature formatting
• Realistic feature distributions
• Balanced fraud representation
• Clean CSV structure

This allows users to focus on analysis, experimentation, and model development rather than extensive data cleaning.

Conclusion

The Telecom Fraud Detection Dataset provides a practical and realistic foundation for building machine learning models that can identify fraudulent calling behavior using behavioral analytics, reputation indicators, and graph-based intelligence. With 10,000 synthetic call records, balanced class distribution, and carefully designed features, it enables users to explore fraud detection techniques, feature engineering, model evaluation, and explainable AI in a privacy-safe environment. Whether you’re a data scientist, researcher, student, or enterprise team, this dataset offers valuable opportunities to develop, benchmark, and improve intelligent fraud detection solutions for modern telecommunications systems.