A Crowdsourced Test Defect Number Prediction Model Based on Test Labor and Test Reports

Abstract

Software reliability growth model is widely used in measurement, prediction and reliability assurance. The uncertainty of software potential defects and the unpredictability of the distributed crowdsourced test process make the crowdsourced test platform crave software reliability modeling techniques to predict the number of potential defects of the software, to evaluate the progress of the test task. This paper puts forward a crowdsourced test defect number prediction model (CTDNPM) that considers both the quantity of test labor and the number of test reports as two test cost elements. A new reliability modeling framework is established based on element correlation, and three existing test work functions are combined to solve the equations, to predict the number of potential defects and the cumulative number of defects detected. The experimental results of four groups of real crowdsourced test data sets show that CTDNPM can predict the number of defects. The error of defect number estimation in the model is less than 10%, which has important guiding significance for monitoring the progress of the test task in the actual crowdsourced test.