Software defects prediction using estimation theory

An early accurate prediction of the number of software defects to be discovered at the end of system testing not only contributes for scheduling and management of resources for software testing but also provides an estimate of product's required maintenance. A composite defects estimation solution is proposed here which provides software defects estimate with reasonably high accuracy at the initial stages of system testing. If only the defects data from system testing is available then a solution based on a Maximum Likelihood Estimator (MLE) is developed. If historical data from past projects is also available a Bayesian estimator is developed to extract prior information and overcome the latency problem that occurs when a long learning period is observed at the initial stages of the system testing phase. A systematic procedure to extract prior knowledge from existing projects is also presented here. Finally Kalman filter is defined for the software defects data to filter the noise incorporated in different phases of software life cycle. It also takes into account the correlation between subsequent defects data points. The solution improves on existing techniques both in terms of latency as well as accuracy.