Empirical assessment of machine learning models for agile software development effort estimation using story points

In the present day developing houses, the procedures adopted during the development of software using agile methodologies are acknowledged as a better option than the procedures followed during conventional software development due to its innate characteristics such as iterative development, rapid delivery and reduced risk. Hence, it is desirable that the software development industries should have proper planning for estimating the effort required in agile software development. The existing techniques such as expert opinion, analogy and disaggregation are mostly observed to be ad hoc and in this manner inclined to be mistaken in a number of cases. One of the various approaches for calculating effort of agile projects in an empirical way is the story point approach (SPA). This paper presents a study on analysis of prediction accuracy of estimation process executed in order to improve it using SPA. Different machine learning techniques such as decision tree, stochastic gradient boosting and random forest are considered in order to assess prediction more qualitatively. A comparative analysis of these techniques with existing techniques is also presented and analyzed in order to critically examine their performance.