Bayesian network-based risk prediction in virtual organizations

Abstract

To support and increase the success rate of collaboration in Virtual Organizations (VOs), usually formed within Virtual organizations Breeding Environments (VBEs), their operation stage and performance of their tasks must be continuously monitored and supervised. A task in the VO is either planned to be performed by an individual partner or jointly by a group of partners, and typically consists of several sub-tasks defining the day-to-day activities of its involved partners. However, VOs are dynamic and therefore detailed activities related to sub-tasks are defined gradually during their operation phase. In this paper, as the base for discovery of potential task failures, past performance and record of previous sub-tasks’ fulfillment of each partner (so-called agent) is considered for appraisal of its trustworthiness. Furthermore, the communication characteristic of the agent and its current workload in all its involved VOs within the VBE are also considered as input for measuring its potential probability of failure on currently assigned sub-tasks. For tasks that involve several partners, a Bayesian network is created during the VO’s operation phase, and used for measuring their failure probabilities. These two potential risk measurements in VOs enable their coordinators to appropriately identify the weak points in their planning of upcoming VO activities, as well as assisting them with advice on how to intervene and change the situation.