Forecasting Software Effort Estimation from UML Class Models Using Predictive Learning

Reasonably accurate Software Development effort prediction is a significant aspect of software product creation. Accurate forecasting of effort results in construction of quality products economically within expected timelines. This work provides a convenient and efficient solution to this issue by taking advantage of the attributes present in the UML Class models of Object-based products. These attributes, along with our custom-built regression analysis python programs were utilized for the estimation forecasting. Our work applied: Linear Regression, Decision Tree Regression, Multilayer Perceptron (MLP) Regressor, Support Vector Regression (SVR), Sequential Minimal Optimization (SMO) with Sigmoid, Polynomial and Radial Basis Function (RBF) kernels to predict the estimation efforts for newly received projects. As per the tests, it was apparent that the SMO model with Polynomial kernel displayed the optimal accuracy having NMSE value being 0.065 and R2 value being 0.935, compared to other regression models.