Designing Machine Learning Method for Software Project Effort Prediction
Ketema Kifle Gebretsadik, Walelign Tewabe Sewunetie · 2019
Software project effort Prediction is the most challenging and important activities in software project development. In software Industry when the size of the project and number of developers increases the project become complex, at this point accuracy prediction is strongly required during the early stages of project development. But to predict at early stages data and information are not available at the preliminary phases of project as well as the data is not complete, consistent and certain. In this research work we uses Artificial Neural network, Fuzzy logic, use case point model, modified environmental factor and revised use case point to predict software project effort at early stages. Artificial Neural Network has the ability to learn from previous data and fuzzy logic deals with uncertainty and also it provides a technique to deal with imprecision and information granularity. Also the modified environmental factor and revised use case point are used to determine the effort at early stage of the project development. In our experiment we compares each models, fitting accuracy using MMRE and PRED (0.25). The fitting accuracy of the models in terms of MMRE for Neuro-Fuzzy-UCBEM is 0.03, Neuro_UCBEM 0.13, Fuzzy-UCBEM 0.12 and UCBEM 0.22 and the fitting accuracy of the models in terms of Pred (0.25) for Neuro-Fuzzy-UCBEM is 1, Neuro_UCBEM 0.93, Fuzzy-UCBEM 0.93 and UCBEM 0.8. So our experiment result shows that Neuro-fuzzy logic model using revised use case point and modified environmental is best out performing model for software project development effort prediction at early stage than another models. Since the Neuro-fuzzy-UCBEM shows low value of MMRE (Mean of Magnitude of Relative Error) and high value prep (0.25) than Neuro-UCBEM, UCBEM and Fuzzy-UCBEM models.