CBN: Combining Cognitive Map and General Bayesian Network to Predict Cost of Software Maintenance Effort

Abstract

Outsourcing of IT/IS service is now becoming a standard protocol for most of companies. As cost related to maintaining IT/IS service increases quite rapidly due to fierce competition in the market and fast changes in customers’ behavior, how to control its cost emerges as the most important problem in the IT/IS service outsourcing industry. It is customary that software maintenance effort (SME) determines cost of IT/IS outsourcing service. Therefore, both IT/IS service providers and demanders have been focused on measuring the cost inflicted by SME, before reaching mutual agreement on price for IT/IS outsourcing service. Problem with this task is that there exist a large number of relevant factors and decision makers ought to be taken all into consideration systematically, which is very hard to do in reality. In this sense, this study proposes a new method called Cognitive Bayesian Network (CBN) in which SME experts first offer a draft causal map for the target SME problem, and the draft cognitive map is translated into the corresponding General Bayesian Network. To determine exact values of conditional probabilities for all of variables and arcs included in CBN, empirical SME data were applied to the CBN. After all CBN showed several merits- (1) it has a flexible structure enough to incorporate relevant variables at any time, and (2) it is capable of producing robust inference results for the given SME problems with rather high accuracy. To prove the validity of the proposed CBN, we interviewed with an expert having more than 15 years of SME experience to draw a draft CBN and than modified it with real SME data, then compared it with other BN models. We found that the performance of CBN is promising and statistically better than other benchmarking BN models.