Software cost estimation of project management is the most challenging task. The cost estimate used to make a careful assessment of the number of resources and the hours needed for software development. Software project managers believe that an inaccurate estimate leads to project failure, which is a significant problem. Due to the continuous and increasing change in software engineering to software development, making an accurate assessment of the software development cost is extremely difficult by using algorithm-based methods including Analogy-Based, Expert-Based and Constructive Cost Model (COCOMO) models. This paper suggests an alternative approach by performing data mining on pre-processed COCOMO NASA benchmark data. This prediction made using three machine learning techniques: Naïve Bayes, Logistic Model Tree (LMT), and Adaptive Boosting (Adaboost). Applied models tested by ten folds cross-validation, and evaluation performed by Precision, Recall, F-Measure, and Matthews correlation coefficient (MCC). All applied techniques have achieved a reasonable error rate. However, the best performance obtained by using LMT and Naïve Bayes. Even though LMT outperformed the other two methods at all levels. The results confirmed the validity of the data mining in general and the technique implemented in particular to the software estimate.
Volume 12 | Issue 5
Pages: 176-185
DOI: 10.5373/JARDCS/V12I5/20201702