The combined model for software development effort estimation using polynomial regression for heterogeneous projects

Abstract

Subject matter: Estimating the software work is a crucial job of persons participating in software project management. The difficulty in predicting effort is compounded by the fact that software development is always changing. In the past, researchers used one form of development methodology in their work to estimate effort and time. Estimations of the software projects are estimated with different size matrices. The lines of code, story point and use case point are required for the estimation using algorithmic models for procedural, agile, and object-oriented development approaches. Currently, the companies use these three types of size matrices for estimating projects. Not any one model present estimates the effort for different development approaches with different size metrics. This paper proposes a combined software estimation model for three types of development methodologies with regression analysis. The estimation can be done with the proposed model for a software project developed using the procedural, agile, and object-oriented approach. Method: The input for the model is the size of the software, such as lines of code, story point, and use case point. The model is developed using the polynomial regression. The model is developed with the four constant parameters that are based on the procedural, agile, and object-oriented projects. A dataset of python projects for procedural, zia dataset for agile, company dataset for object-oriented methodology is used to propose the model. Conclusion: The effort is predicted for the procedural, agile, and object-oriented projects with the polynomial regression model and compare the results to existing models to validate the work. The R2 is used to measure accuracy and the MMRE is used to determine error. The accuracy of the proposed model was higher than 90% and the error was found to be less than 0.05. The results are compared with case-based reasoning and an ensemble model for the procedural approach, linear regression and Bayesian network for the agile approach, and linear and log-linear regression for object-oriented approach. The minimum error and maximum accuracy is achieved compared to these techniques.