Developing a Team Performance Prediction Model: A Rough Sets Approach

Abstract

This paper presents a rough sets approach in developing a team performance prediction model. To establish the prediction model, a dataset from an academic setting was obtained, consisting of four predictor variables: prior academic achievement, personality types, personality diversity, and software development methodology. In this study, four main steps in rough set, including discretisation, reduct generation rules, generation of decision rules, and evaluation were used to develop the prediction model. Two reduction algorithms; a genetic algorithm (GA) and a Johnson algorithm were used to obtain optimal classification accuracy. Results show that the Johnson algorithm outperformed the GA with 78.33% model prediction accuracy when using 10-fold cross validation. The result clearly shows that the rough sets is able to uncover complex factors in team dynamism, which revealed that the combination of the four predictor variables are important in developing the team performance prediction model. The model provides a practical contribution in predicting team performance.