Optimizing Detection Thresholds in Anomaly Prediction Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods in semiconductor manufacturing often fail to accurately predict component defects during pre-tests, leading to delayed identification of issues and significant production waste, as they rely solely on anomaly scores without considering explainability values, resulting in incorrect classifications and increased rejection rates.
Innovation Solution
A computer-implemented method that optimizes the detection threshold of a prediction model by incorporating anomaly scores and explainability values, such as Shapley values, to improve predictive accuracy and identify contributing parameters, allowing for early detection of process problems and reduced waste.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If anomaly detection relies solely on anomaly scores without explainability values, then the detection process is simple and fast, but predictive accuracy decreases and incorrect classifications increase
Solution Approach 1:
The patent combines anomaly scores and explainability values into a unified detection framework. The detection threshold is optimized by integrating both types of information, allowing the system to leverage the simplicity of anomaly scores while incorporating the interpretive power of explainability values to improve predictive accuracy and reduce incorrect classifications.
Solution Approach 2:
The patent optimizes the detection threshold as a可调 parameter by analyzing the relationship between anomaly scores and explainability values. This parameter optimization enables the system to adapt to different operating conditions and maximize predictive accuracy while maintaining a balance between detection sensitivity and false positive rates.
2Reliability
If the detection threshold is set to be highly sensitive, then more anomalies are detected, but false positive rate increases and rejection rate increases
Solution Approach 1:
The patent employs feedback mechanisms where the detection results, including both anomaly scores and explainability values, are used to continuously optimize the detection threshold. This feedback loop allows the system to learn from past decisions and adjust the threshold to minimize false positives while maintaining high anomaly detection reliability, thereby reducing unnecessary rejections and production waste.
Solution Approach 2:
The detection threshold is not fixed but dynamically adjusted based on the combined information from anomaly scores and explainability values. This dynamic adjustment enables the system to adapt to varying operational conditions and maintain optimal performance, balancing sensitivity and specificity to reduce both missed anomalies and false alarms.
3Loss of substance
If pre-test evaluation is performed to detect anomalies early, then production waste is reduced, but detection accuracy must be sufficient to avoid premature rejection
Solution Approach 1:
The patent performs preliminary anomaly detection during the pre-test evaluation phase by analyzing both anomaly scores and explainability values. This early detection capability allows the system to identify potential anomalies before they lead to defective products, enabling early intervention and reducing production waste while maintaining sufficient detection accuracy through the combined use of multiple evaluation metrics.
Data Source
AI summary
A computer-implemented method for optimizing a detection threshold of a prediction model used to determine an anomaly of a component is disclosed. The detection threshold indicates the criterion above which the prediction model classifies a component as anomalous. The method includes (i) providing the prediction model, (ii) providing a plurality of pre-test results determined for a plurality of test parameters for a plurality of components, respectively, (iii) providing a final test result for each of the plurality of components, wherein the respective final test result indicates whether the respective component has an anomaly in a final test, (iv) calculating a respective anomaly result for each of the plurality of components by evaluating the respective pre-test results by the prediction model, (v) calculating a respective explainability value for each of the plurality of test parameters for each of the plurality of components by the prediction model, and (vi) optimizing the detection threshold based on the calculated anomaly results, the calculated explainability values and the final test results provided.