The Fraud Control Project is a comprehensive solution for analyzing and scoring SMS messages for potential fraud or spam risks. It consists of two primary components:
The Fraud Control API is a web service built using .NET 8.0, designed to provide a robust fraud detection mechanism by integrating various AI models and external services.
- GeminiAI: This service uses a machine learning model to evaluate SMS messages and provide a risk score. The score indicates the likelihood of the message being fraudulent or spammy.
- TensorFlow Model: This model processes SMS messages with TensorFlow to generate a risk score based on its trained machine learning model. The score is standardized to a 0 to 100 range.
- IPQS: This service checks any URLs present in the SMS message against a database of known malicious IP addresses and URLs. It provides a score or -1 if no URL is found.
- OpenAI: This service uses advanced language models from OpenAI to analyze the content of SMS messages and determine if they are likely to be fraudulent or spammy.
The final fraud score is a weighted combination of scores from the above services:
- GeminiAI Score: 30%
- TensorFlow Score: 20%
- IPQS Score: 10% (only if a URL is present, otherwise -1)
- OpenAI Score: 40%
The final score is calculated as:
- With IPQS Score:
Final Score = (GeminiAI Score * 0.3) + (TensorFlow Score * 0.2) + (IPQS Score * 0.1) + (OpenAI Score * 0.4) - Without IPQS Score:
Final Score = (GeminiAI Score * 0.3) + (TensorFlow Score * 0.3) + (OpenAI Score * 0.5)
Score Ranges and Risk Levels:
- High Risk: Final Score >= 80
- Moderate Risk: 50 <= Final Score < 80
- Low Risk: Final Score < 50
Fraud Control for SMS is a standalone tool designed to detect fraudulent SMS messages using various AI models and external APIs.
- OpenAI Model: This model analyzes SMS content using OpenAI's language models to detect any suspicious patterns and provides a fraud probability score.
- Gemini AI Model: Similar to OpenAI, this model evaluates SMS messages for potential fraud using a machine learning model and provides a risk score.
- TensorFlow Model: This model uses TensorFlow to determine if an SMS is spam, based on its trained dataset, and provides a spam probability score.
- IPQS API: This API assesses URLs included in SMS messages for potential threats such as phishing or malware and provides a fraud probability score.
The fraud detection program aggregates the scores from the Gemini AI model, TensorFlow model, and IPQS API to generate a final risk score. This score is a weighted average of the individual scores, giving a comprehensive assessment of the SMS message’s risk level.