Debashish, Intel Corporation
Many organizations want to estimate the number of defects prior to product’s launch. This helps them to estimate the quality of product, maintenance efforts, and risk in meeting schedule. Project management teams struggle to define milestone timelines and understand if Alpha, Beta, and PV (Production) dates are too early or too late and how the number of open defects impacts the program schedule.
To address the above concerns we created a defect prediction model based on empirical data analysis. This Model helps in understanding the impact of design and testing processes on defect counts and failure and
helps in planning milestones (total defects) and also in process monitoring of defect discovery numbers (latent defects).
In this model, we use the historical data from previous projects and for the new projects we perform delta analysis. After modeling for 30 projects we observed that the plot (Cumulative Defects wrt Work week) follows a non-linear sigmoid pattern. Defects with all severities including duplicates/rejects and deferred defects are used for this model. To find the parameters in the sigmoidal pattern we used the statistical tool JMP (from SAS). We found that the Logistical 3P model turns out to be the most accurate with a fit value R2 (statistical measure of how close the data is to the fitted regression line) and having error rate less than 0.01%. The 3P refers to 3 parameters: Inflection Point (week where 50% of total # of defects are found), Asymptote (Total # of defects) and Growth Rate (Rate at which defects grows each week). Delta analysis consists of any changes wrt features, project complexity, resources etc. This defect prediction model can aid the project management teams to map the SW milestones and impact of total number of defects on the project schedule and avoid any schedule delays and milestone commitments.
