VoIP Illegal Call Detection Using CNN Image Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
VoIP systems are vulnerable to illegal calls due to unchanging default IDs and passwords, leading to significant billing damages as existing fraud detection methods are inadequate in real-time prevention.
Innovation Solution
An apparatus and method using a CNN algorithm to detect illegal call patterns by converting call data into images, classifying them as normal or illegal, and blocking the subscriber terminal upon detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If default IDs and passwords are used in enterprise IP-PBXs without changing them, then device complexity is reduced and ease of operation is improved, but security vulnerability increases leading to illegal calls
Solution Approach 1:
The system performs preliminary analysis of call patterns and user behavior before illegal calls occur. By establishing baseline behavior profiles during normal operation, the system can detect deviations that indicate unauthorized access or fraudulent activity, preventing damage before it occurs.
Solution Approach 2:
The patent introduces an intermediary analysis system that sits between the VoIP infrastructure and the billing system. This intermediary layer analyzes call detail records and packet data to detect fraudulent patterns, acting as a mediator that prevents illegal calls from reaching the billing system without requiring changes to the underlying IP-PBX security configuration.
2Measurement precision
If existing fraud detection methods are used, then device complexity is kept low, but detection precision is insufficient leading to billing damages
Solution Approach 1:
The patent replaces traditional rule-based fraud detection mechanisms with a machine learning-based analysis system. By using artificial intelligence to analyze call patterns, the system achieves superior detection precision without requiring complex manual rule configuration, automatically adapting to new fraud patterns through continuous learning.
Solution Approach 2:
The system dynamically changes analysis parameters based on learned patterns from training data. By adjusting detection thresholds, analysis depth, and pattern recognition parameters based on historical fraud data and current call patterns, the system optimizes detection precision while maintaining operational efficiency.
3Reliability
If real-time detection of illegal calls is implemented, then billing damage prevention is improved, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary training offline using historical fraud data to pre-compute detection models and patterns. During real-time operation, the pre-trained models quickly classify incoming calls without requiring extensive computational resources, achieving both high reliability and fast processing speed.
Solution Approach 2:
The patent implements a two-stage detection process where only calls with certain risk indicators undergo full analysis. By applying partial analysis to low-risk calls and excessive (full) analysis only when necessary, the system achieves high detection reliability while minimizing overall processing time and resource consumption.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An illegal call detection apparatus comprises a data collecting unit, a preprocessing unit, and a learning unit. The data collecting unit collects, from at least one of a subscriber terminal or a call exchanger, a raw packet generated by the subscriber terminal using a VoIP service, and collects, from the billing server, CDR data related to the raw packet. The preprocessing unit generates learning data by using service usage information extracted from the CDR data and service detailed information extracted from the raw packet, and generates a training image by converting the training data into an image according to a predetermined imaging rule. The learning unit extracts at least one or more features from the training image, and learns whether the training image is related to an illegal call by using the features, through an illegal call detection model.