Software Development Project Efforts and Time Forecasting Based on a Neural Network

EasyChair Preprint 1788

Abstract

The problems of forecasting labor costs and lead time for software development projects remains relevant, despite a large number of studies in this area. There are options for solving it, both using various analytical descriptions (formulas) with empirical coefficients, and proposals for using various types of predictive models. Of the analytical solutions to this problem, the most famous is COCOMO II.
The hypothesis of this study is the assumption that it is possible to develop a methodology for predicting labor costs and project lead time based on the use of a trained neural network with the results no worse than COCOMO II gives, but without the need to expertly determine and calculate the coefficients of model factors leading to the COCOMO II calculation formula.
To solve the problem in the research, variables are defined, criteria and metrics are determined for the forecasting model, a neural network training method is selected, a neural network is created and trained, experimentation is made, research of the developed forecasting model is carried out, analysis and evaluation of the results obtained.
In this study, it is planned to experimentally prove that the result of the developed forecasting methodology will be no worse than the forecast of the labor costs of developing the project implementation time obtained using the COCOMO II model.
To validate the developed forecasting model, it is planned to use expert assessment as well.
As a result of the research work, a methodology for predicting labor costs and software development time based on neural networks will be obtained. The novelty of the described study is the ability to automate the tuning of the prediction apparatus (neural network) to the metric basis of a specific project instead of performing expert determination of factor coefficients for forecasting using the COCOMO II model.

Keyphrases: CMMI, COCOMO II, IT project, model training

@booklet{EasyChair:1788,
  author    = {Konstantīns Dinārs},
  title     = {Software Development Project Efforts and Time Forecasting Based on a Neural Network},
  howpublished = {EasyChair Preprint 1788},
  year      = {EasyChair, 2019}}