Reliability Assurance in Early-Life-Failure Test through Improved Nearest Neighbor Regression

Abstract

During manufacturing test, researchers usually overlook the importance of process variation defects and marginal defects, which can seriously affect test results of Early-Life-Failure (ELF). Theoretically, machine learning classification methods can be used to identify these two defects. In fact, when features overfitting or data distribution overlap seriously, classifiers perform poorly, it will not achieve the desired results. This paper first-ever proposes a kind of data preprocessing method combines improved K-Nearest Neighbors (KNN) regression classifier, so that the classification results will be enhanced in terms of classification performance. Experiment results demonstrate that the predictive accuracy is 45% higher than the traditional logistic regression method. This proposed method will drive critical new requirements for fault modeling, test generation and test application, and implementing them effectively will require a new level of collaboration between process and product developers.