Use of Discriminant Analysis in Forecasting the Success of a Software Development Project

Abstract

Cost overruns are widespread in software development projects. Though various tools for risk analysis are in use, a manager still needs a simple and powerful method predicting success or failure of software projects. This paper proposes discriminant analysis as an appropriate tool. It has successfully been used for predicting bancruptcy and for credit contract classification. Discriminant analysis needs predictor variables for assigning a group classification (successful / failing) to each project. The data base for predictors and groups was established in a major German company by analyzing 107 software projects. The discriminant functions calculated can be used as a caveat against cost overruns as the reliability of failure predictions could be proven to be sufficiently high.