A Machine Learning Attempt for Anatomizing Software Risks in Small and Medium Agile Enterprises

Abstract

The ultimate aim of customer satisfaction and the increasing number of unexpected risks in a changing Agile Software Development (ASD) environment, one of the most important rising demands is in the area of systematic but light-weight risk management tools and methodologies. Risk analysis is a significant phase in the process of risk assessment, which helps to evaluate the risks in order to mitigate them effectively within a limited duration. Recently, machine learning algorithms have become popular for solving problems in various domains, including software risk analysis and prioritization, due to their better performance and efficiency. With this aspect, an approach for predicting the level of software risks with the proposed risk dataset has been attempted in this study with the basic machine learning algorithms for risk classification purposes. The logistic regression, decision tree, Support Vector Machine (SVM), naïve bayes, and K-Nearest Neighbor (KNN) algorithms were implemented in the experimental analysis. The results reveal that the proposed dataset renders better outcomes with logistic regression (70% accuracy) and SVM (65% accuracy). Out of the five algorithms, the exclusion of the Agile Software Risk Identification (ASRI) framework attribute ‘Risk Nature’ from the overall proposed risk dataset has a more negative impact on the performance of the logistic regression, decision tree, and KNN models than the exclusion of the Goal-driven Software development Risk Management (GSRM) framework attribute ‘performance goal affected’. This indicates that the ‘Risk Nature’ attribute plays a significant role in analyzing the risks and predicting their level of importance.