A Software Effort Estimation Method Based On Fuzzy Decision Tree

Software estimation is a key component of software project management, that's because the success of software projects relies on the accuracy of software estimation. As software development is a gradual process, so our understanding of the characteristics collaborated with the main factors affecting the progress of the projects become clear gradually; in addition, our cognitive of those various factors that influencing the accuracy of software estimate is also a gradual process. Traditional estimation techniques lack the ability to deal well with qualitative data and the inexplicit but known on the common sense data.Fuzzy technology can handle the dynamic problems which are complex, nonlinear information. Decision tree, a kind of instance-based inductive learning algorithm, is intuitive and easy to understand. The decision tree techniques collaborated with fuzzy logic, fuzzy decision tree, can integrate the advantages of both technologies. This paper conducted the software effort estimation using this method to identify the operation law and performance of research estimates.Firstly, this paper introduced the fuzzy sets, fuzzy logic, the concept of fuzzy decision trees, and further discussed in generating a fuzzy decision tree. Subsequently, evaluation criteria and standards in software estimation model performance were reviewed, and the general process of estimation methods with the fuzzy decision tree basing on ID3 was described. Finally, the estimation model basing on Desharnais dataset was explored. This study focused on the effects of the criterion to stop splitting the fuzzy decision tree node and de-fuzzication methods upon the performance of the model.This study presented the distinctive effects of the criterion to stop splitting the node of fuzzy decision tree on the performance of the model, but depended on various defuzzification methods. This study also found that fuzzy decision tree in accordance with the people's thinking habits of linguistic expression, which could be understood and accepted easily for the developing and prediction process. However, compared to the classical decision tree, fuzzy decision trees have some disadvantages. For example, fuzzing, defuzzifying, the selection of decision attributes and the partitioning computational for nods are more complexity; in addition, those samples could be attached to multiple nodes, then it made the fuzzy decision tree structure more complicated than classical decision tree.