Proactive detection of voice phishing networks using call log analysis and machine learning.

Area of Science:

• Cybersecurity
• Data Science
• Financial Criminology

Background:

• Voice phishing is a growing cyber-enabled financial fraud exploiting telecommunications.
• Reactive policing struggles to prevent real-time voice phishing incidents.
• A proactive detection method is needed to combat telecommunication fraud.

Purpose of the Study:

• To introduce a data-driven framework for proactive detection of voice phishing phone numbers.
• To analyze behavioral patterns in call log data to differentiate fraudulent from legitimate users.
• To develop and evaluate machine learning models for classifying suspicious phone numbers.

Main Methods:

• Utilized large-scale call log data from South Korea.
• Extracted behavioral features from call metadata.
• Employed stepwise logistic regression, random forest, gradient boosting, and Multi-Layer Perceptron (MLP) models for classification.

Main Results:

• Voice phishing numbers exhibit distinct behavioral patterns: concentrated weekday activity, shorter calls, higher outgoing ratios, and MVNO preference.
• Machine learning models achieved high performance: >95% accuracy and >97% recall.
• The framework effectively identifies suspicious numbers linked to voice phishing operations.

Conclusions:

• AI and behavioral analytics can be integrated into proactive fraud detection systems.
• The study provides a robust method for early-warning mechanisms against telecommunication-based financial crimes.
• This data-driven approach enhances the prevention of voice phishing fraud.