Software Effort Estimation for Agile Software Development Using a Strategy Based on k-Nearest Neighbors Algorithm

Abstract

Agile development adoption in organizations is a trend that continues to accelerate according to the 15th State of Agile Report. Enterprises need to respond quickly to the needs of their customers and stakeholders and by adopting agile practices in IT teams, business value is raised in both performance and quality, so it is important to adopt practices and models that ensure the time, scope and cost of a project are achieved successfully. This paper presents a hybrid effort estimation model that uses a story point approach with machine learning techniques to estimate completion time and total cost of a project that is developed with agile methods like Scrum. The main machine learning technique used to implement the project is the k-Nearest Neighbors algorithm (KNN), its learning capabilities are assessed through 10-Fold cross validation and the estimates are compared with the original dataset and the results obtained from literature to show that estimates are competitive. The proposed approach uses category size labels that improve the original estimation model based on linear regression. The research uses 21 projects developed by six software houses, and training is done on a set created from a technique called data augmentation that generates 42 projects with a small amount of noise. Completion time is measured in days and total cost is valued in Pakistan rupees. All the results are evaluated through accuracy, Mean Squared Error, Mean Relative Error, variance and coefficient of determination.