Classification of Group Potency Levels of Software Development Student Teams

Abstract

"This paper describes the use of an automaticclassifier to model group potency levels within softwaredevelopment projects. A set of machine learning experimentsthat looked at different group characteristics and variouscollaboration measures extracted from a team’s communicationactivities were used to predict overall group potency levels.These textual communication exchanges were collected fromthree software development projects involving students livingin the US, Turkey and Panama. Based on the group potencyliterature, group-level measures such as skill diversity, cohesion,and collaboration were developed and then collected for eachteam. A regression analysis was originally performed on thecontinuous group potency values to test the relationships betweenthe group-level measures and group potency levels. This method,however, proved to be ineffective. As a result, the group potencyvalues were converted into binary labels and the relationshipsbetween the group-level measures and group potency werere-analyzed using machine learning classifiers. Results of thisnew analysis indicated an improvement in the accuracy of themodel. Thus, we were able to successfully characterize teamsas having either low or high potency levels. Such informationcan prove useful to both managers and leaders of teams in anysetting."