Software Defect Prediction Based on Stability Test Data

Software defect prediction is an essential part of evaluating product readiness in terms of software quality prior to the software delivery. As a new software load with new features and bug fixes becomes available, stability tests are performed typically with a call load generator in a full configuration environment. Defect data from the stability test provides most accurate information required for the software quality assessment. This paper presents a software defect prediction model using defect data from stability test. We demonstrate that test run duration in hours is a better measure than calendar time in days for predicting the number of defects in a software release. An exponential reliability growth model is applied to the defect data with respect to test run duration. We then address how to identify whether estimates of the model parameters are stable enough for assuring the prediction accuracy.