Genetic algorithm based probabilistic model for agile project success in global software development

Mohammad Shameem

doi:10.1016/j.asoc.2023.109998

2023-03-01

PRE

Mohammad Shameem *

Mohammad Nadeem

Abu Taha Zamani

DOI：10.1016/j.asoc.2023.109998

原文链接

原文求助

摘要

En 中文

Software development organizations implement agile methods in global software development (GSD) to leverage benefits in terms of low development cost, continuous project delivery, and high-quality product. Despite the benefits, using the agile process in the GSD organizations is not a straightforward task due to inhibit additional risk factors that could lead agile projects to failure. GSD organizations are continuously improving their process management activities to improve the rate of agile project success. Project success could be improved if the project manager or organization management has preliminary information about the derailment of the project features. This work aims to investigate the most influential agile project features in determining the project outcomes and develop a costeffective, and effort-based prediction model to improve the probability of successful completion in the globally distributed environment. To do so, a nature-inspired optimization algorithm i.e., genetic algorithm (GA), has been employed to achieve these goals. In the proposed model, GA considers the probability of success with respect to cost to determine probable project outcomes. An efficacy measure is formulated as a fitness function in GA which maximizes the success of agile project outcomes relative to cost. The optimization model has been tested with two different prediction models i.e., Naive Bayes classifier (NBC) and logistic regression (LR). We performed the experiment on data gathered through the survey administered from the globally distributed agile projects. The results demonstrate that prediction models calculate the efficacy for best solutions as 0.531 and 0.5850 for NBC and LR, respectively. Moreover, the ranking of each project feature based on their relative cost identified using NBC and LR have more similarities. The t test results are significant, i.e., t = 6.068, p = 0.001 < 0.005, which indicates that no significant differences have been observed between the ranking assigned by two different methods ( NBC and LR). The results reveal that the developed prediction model based on identified eight agile project features that the GSD organization management and agile team need to focus more on to facilitate cost-effective successful implementation of agile projects. (c) 2023 Elsevier B.V. All rights reserved.

Keyword:

Agile development

GSD

Success probability

Prediction

Genetic algorithm