Discovering Many-to-One Causality in Software Project Risk Analysis

Abstract

Many risk factors affect software development and risk management has become one of the major activities in software development. Discovering causal directions among risk factors and project performance are important support for risk management. The Additive Noise Model (ANM) is an effective algorithm for discovering the direction on one-to-one causalities, but ineffective on many-to-one causalities which are frequent in software project risk analysis (SPRA) process. Thus we proposed a modified ANM with Conditional Probability Table (ANMCPT) to discover the causal direction among risk factors and project performance. The experimental results show our proposed algorithm is effective to discover the many-to-one causalities in SPRM on 498 collected software project data, and it performs better than other algorithms in the prediction with discovered causes of project performance, such as logistic regression, C4.5, Naïve Bayes, and general BNs. This study firstly presents an approach using ANM for many-to-one causality discovery in SPRA and then proves that it is an effective algorithm for analyzing the risk in software project.