From data mining to decision support using qualitative multi-criteria models
Data is an excellent source of information for decision support. Viewed as a historical record of past decisions, data can be analyzed in order to understand those decisions, to find out the main principles that governed them, and to develop models that can guide or predict future decisions. Unfortunately, models developed by machine learning algorithms from data often turn out not to be fully suitable for decision support, because they are, for instance, incomplete or inconsistent. To make them suitable for decision support, such models have to be verified, revised and/or extended by human experts, with the aim to improve their completeness, consistency and comprehensibility.
In this talk, I will present an approach based on a qualitative, rule-based, hierarchical multi-criteria modelling method DEX. In the context of using data for decision support, this approach facilitates combining knowledge from different sources (such as data mining and expert modeling) and supports fulfilling the desired quality criteria for decision models. I will use real cases from the areas of food production, risk assessment and health-care management, to show: (1) the inadequacy of models developed only from data, (2) the need for their improvement through expert modelling, and (3) possible ways of adapting them for decision support through hierarchical multi-criteria models.
About the lecturer
Marko Bohanec is a leading Slovenian researcher in the field of decision support models and systems. He works at the Department of Knowledge Technologies at the Jožef Stefan Institute. Also, he is a professor of computer science at the University of Nova Gorica and a visiting professor at the University of Belgrade. He is one of the pioneers of qualitative multi-attribute modeling, which combines the fields of decision analysis and artificial intelligence to support people in making complex decisions. He is a co-author of the method DEX and computer program DEXi, which are used for decision support nationally and internationally. In the last ten years, he has been involved as a decision support expert in European projects on the production, distribution and quality control of food and feed (projects ECOGEN, SIGMEA, Co-Extra and DECATHLON), and disease management in health-care (PD_manager and HeartMan). He has co-authored more than 60 journal papers. According to Google Scholar his work has been cited about 2500 times.