Forecasting software development costs in scrum iterations using ordinary least squares method

Abstract

During scrum iterations, it is possible to apply cost forecasting for software testing and operation, if the data from previous iterations are known. Since the data for estimating the scope of work and the deadline within one sprint are accumulated during the project execution, it is possible to use such data to build a forecasting algorithm for the estimated parameters of the subsequent sprints.

The approach is based on refining the assessment provided by the development team and the scrum master in a specific metric. The main parameters for evaluation are the execution time and the amount of work performed. As a result of forecasting, it is possible to obtain clarifications for the team's assessment regarding the scope of work for the next sprint. This estimate is based on planned and actual data from the previous sprints.

The article discusses the method of least squares and the proposed code for a machine learning model based on this method. An example and graphs for iterations in scrum and corresponding forecasting for the next sprints are presented.

The use of the least squares method allows creating a mathematical model that can be adapted to different project conditions, providing flexibility and accuracy in forecasting. For example, the study uses the real data from the previous sprints, which includes the team's resource assessment and actual expenditures. Based on these data, a model was built that demonstrates a high correlation between predicted and actual costs, confirming the effectiveness of using the least squares method.

So, the least squares method is an effective tool for forecasting software development costs in scrum iterations. This method allows development teams to better plan their resources and timelines, contributing to the overall efficiency of the project.