Enhanced testing performance via unbiased test sets

The test generation task involves two separate questions which are: (1) What should the next test be? and (2) Have enough tests been selected to achieve an acceptable defective part level? Historically, the same fault set (usually the stuck-at-fault set) has been used to answer both questions. When both questions use the same fault set, a statistical bias is introduced to the answer of the second question. In this paper, we propose the use of independent models for answers to the two questions above, and we show, via probabilistic analysis as well as experiments, that the result is a superior test set selection method.<
>