Fraudulent Traffic Detection via Distribution Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting fraudulent web traffic in online advertising are inefficient, as they rely on binary matching mechanisms that burden systems with the need to check each incoming signal, leading to significant waste in advertising spending due to illegitimate traffic.
Innovation Solution
A computer-implemented method that compares an empirical distribution of web traffic features to a model distribution of reliable traffic, determining a minimum number of datapoints to remove to create a modified distribution within a significance level, and generates an alert if the difference exceeds a threshold, indicating fraudulent traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If binary matching mechanisms are used to detect fraudulent traffic by checking each incoming signal against blacklisted signatures, then fraudulent traffic can be blocked, but the system becomes burdensome and inefficient due to the need to check every incoming signal
Solution Approach 1:
The system pre-calculates and stores baseline traffic patterns for webpages and creatives before fraudulent traffic arrives. By having these reference distributions ready in advance, the system can quickly compare incoming traffic against pre-established baselines using statistical tests, eliminating the need to analyze each individual signal in real-time while maintaining reliable fraud detection
2Measurement precision
If baseline traffic behavior is established for webpages or creatives to detect significant changes, then fraudulent traffic detection can be improved, but the complexity of the detection system increases
Solution Approach 1:
The system transforms the detection approach by changing from analyzing individual signal characteristics to analyzing statistical distribution parameters of traffic patterns. By using statistical tests that compare distribution parameters (such as frequency distributions of traffic features) between baseline and current traffic, the system achieves high detection precision while maintaining manageable complexity through mathematical abstraction
3Measurement precision
If statistical methods are used to compare traffic distributions and determine significance levels, then detection accuracy improves, but the computational requirements increase
Solution Approach 1:
The system extracts only the essential statistical features and distribution parameters from traffic data that are necessary for fraud detection, rather than processing complete traffic signals. By focusing computational resources on calculating and comparing key distribution parameters through statistical tests, the system achieves high detection accuracy while minimizing computational energy consumption by eliminating unnecessary data processing
Data Source
AI summary
Methods and systems for fraudulent traffic detection and estimation are disclosed. Initially, an empirical distribution of a plurality of features based on a first plurality of datapoints for the plurality of features is received. Next, a model distribution of the plurality of features based on a second plurality of datapoints for the plurality of features is received. Then, it is determined, a minimum number of datapoints to remove from the first plurality of datapoints to create a modified empirical distribution corresponding to the model distribution within a first significance level. Finally, an alert that the first plurality of web traffic includes at least one fraudulent instance of web traffic is generated in response to the determination that the minimum number of datapoints is greater than a first threshold.