|M.Sc Student||Sholokhman Dmitry|
|Subject||A Comparison of Methods for Evaluation of Quality of Code|
|Department||Department of Quality Assurance and Reliability||Supervisor||Professor Eliezer Kantorowitz|
Comparison a Methods for Evaluating the Quality of Code Inspection.
In order to control software inspection, it must be decided whether an inspected document is of sufficient quality and can be passed to the next development step. To make this decision the number of remaining defects in the document would be useful. Unfortunately the number of faults, before or after inspection is unknown and we need to estimate it in some way. For this purpose statistical models were developed in the software engineering field. These models estimate in different ways the number of remaining faults after inspection, based on the results of the inspection. One of these models was developed in Technion by Prof. E.Kantarowitz and is called the Linear Model. This model is described in the work of Arzy Laor and later work of E.Kantarowitz .
I assess the performance of the linear model estimates under the following two criteria.
1. The central tendency of the estimators’ accuracy. This can be used to describe the average performance of an estimator.
2. The variability of estimators’ accuracy. The variability relates to repeatability of an estimator.
In this work the number of defects is estimated and compared to the actual number of defects. Relative error (RE) is used to quantify how good or how poor the estimate is. The RE is define as
RE=(estimated number of defects - real number of defects) / real number of defects.
Important results of the evaluation:
· The accuracy of linear model improves with an increasing number of inspectors.
· Linear model performs better in terms of median relative error than other models do. Unfortunately, it shows very poor behavior in terms of variability, especially for small numbers of inspectors.
· In particular situations described in 8.3 the linear model produces extreme overestimations.