Co-insights for multi-criteria decision making
Abstract
Modern industrial production systems demand the ability to perceive the interrelationships of presented facts in various, complicated datasets, collaboratively to guide action towards a desired objective. With the support of historical, current, and predictive aspects and benefiting from tools of analytics, data mining, business performance management, benchmarking, and predictive analytics, better performance in strategically decision making is tried to be achieved. In production systems, one such instance is the inventory and sales management, where collaboration and insight exchange of production and sales department is necessary on diverse datasets, such as demand patterns, inventory levels, risk of lost sales, inventory quality, and control charts (i.e. EWMA), etc. Collaboration and decision support is essential to make such a collaboration system scalable. In this work, a “Human in the Loop” type collaboration framework is introduced with automated collaboration support features to enable the exchange of insights (Co-Insights Framework). The main contributions from this work are the development of a Coordination Network (Co-Net) model for Co-Insights, defining the task and participant models for Co-Insights; a Collaboration Resource Allocation Analysis mechanism for Task – Participant Matching; a Participant – Interface Matching mechanism and development of a prototype for a Collaborative Visual Analytics (Co-VA) Workspace (with PHP and MySQL based on HUBzero). Thus, by better matching tasks and tools to participants and enabling effective exchange of insights, improvement in multi-criteria decision making process is intended. In the experiments, when comparing proposed Collaboration Resource Allocation Analysis (CRAA) with greedy selection algorithm (GrSA) and random selection algorithm (RSA), CRAA yielded up to 35% improvements in the introduced relevance measure for the Task – Participant Matching. The proposed neural network based participant – interface matching algorithm yielded robust results with 4% deviations for 10% noise levels and with 16% deviations for 30% noise levels.
Degree
M.S.
Advisors
Nof, Purdue University.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.