Improvement of COCOMO II Model to Increase the Accuracy of Effort Estimation

摘要

Effort estimation is one of the most important activities in the software development project, because the project managers need to bea able to estimate the amount of cost and time for developing software. There are several techniques and models that can be used to estimate effort such as COCOMO, COCOMO II, SLIM, SEER-SEM. One method that is widely used for effort estimation is COCOMO II. COCOMO II model is an improvement of COCOMO '81 model. However, COCOMO II estimation results are still not satisfying in terms of accuracy. To improve the accuracy of COCOMO II, many researchers are trying to combine COCOMO II with other methods. In this paper, we propose to combine COCOMO II with K-Means clustering method to improve the accuracy. K-Means clustering is used to determine the data that will be used in the COCOMO II calibration process. The COCOMO II calibration aims to determine the new A and B constant values based on software project dataset. Based on the results of the study, it can be concluded that the accuracy of the proposed method generally increased compared to the original COCOMO II model. The value of accuracy depends on the preprocessing technique performed and the number of clusters. The best accuracy is achieved when the preprocessing technique used is to multiply cost driver attributes by 100 and number of clusters is 5. This proposed method can reduce the value of MRE COCOMO II from 1.32 to 0.85 and increase the value of PRED (0.3) from 32% to 54% for COCOMO NASA 2 dataset and Turkish Software Industry.