Smoothing Risk Values Across Time Categories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning predictive models, particularly in fraud detection and click-through-rate prediction, are vulnerable to minor changes in feature values causing substantial changes in predictive scores, making them susceptible to manipulation by adverse parties.
Innovation Solution
A system and method that involves receiving event data, grouping events into categories based on time ranges, determining initial and secondary risk measures, and applying a smoothing function to reduce discrepancies between adjacent categories, thereby stabilizing predictive scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature encoding is applied to transform raw features for predictive modeling, then the model can process and analyze features effectively, but small changes to feature values cause substantial changes in predictive scores, making the model vulnerable to manipulation
Solution Approach 1:
The patent segments continuous feature values into discrete bins or categories. By dividing the continuous feature space into discrete intervals, small variations in feature values within the same bin do not cause changes in the encoded representation, thereby stabilizing predictive scores while maintaining the ability to distinguish between different feature ranges.
Solution Approach 2:
The patent transforms the original feature values through parameter changes including binning, scaling, and encoding. These transformations modify the parameter representation of features to reduce the impact of minor value changes, converting sensitive continuous parameters into more stable discrete or normalized parameters that maintain discriminatory power without excessive sensitivity.
2Stability of the object's composition
If discrete binning is used to categorize features, then predictive scores become more stable, but the model loses some granularity and precision in distinguishing between closely spaced feature values
Solution Approach 1:
The patent applies different binning strategies and encoding methods to different regions or types of features based on their local characteristics. High-cardinality features with many distinct values receive different treatment compared to low-cardinality features, allowing the system to maintain precision where needed while achieving stability where appropriate.
Solution Approach 2:
The patent employs dynamic binning where bin boundaries and structures are determined based on the data distribution and can adapt to different datasets or changing conditions. This dynamic approach allows the system to optimize the balance between stability and precision by adjusting bin configurations based on the specific characteristics of the input data.
Data Source
AI summary
The present teaching generally relates to removing perturbations from predictive scoring. In one embodiment, data representing a plurality of events detected by a content provider may be received, the data indicating a time that a corresponding event occurred and whether the corresponding event was fraudulent. First category data may be generated by grouping each event into one of a number of categories, each category being associated with a range of times. A first measure of risk for each category may be determined, where the first measure of risk indicates a likelihood that a future event occurring at a future time is fraudulent. Second category data may be generated by processing the first category data and a second measure of risk for each category may be determined. Measure data representing the second measure of risk for each category and the range of times associated with that category may be stored.