Hybrid methods for uncertain knowledge inference : automatic generation of uncertain rules from data
Loading...
Date
2024-11-07
Authors
Κόβας, Κωνσταντίνος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Making decisions or predictions from data through machine learning (ML) methods has been a research topic of high interest in recent years. A disadvantage of these methods is the lack of transparency in the way results are derived, and therefore the inability to provide explanations, because the ML models used do not inherently support this. An alternative approach is to create models that have these capabilities, such as rule-based or case-based models. Furthermore, for most problems there are either incomplete data or the knowledge represented by the data is inherently uncertain, which present challenges that must also be addressed. Another potential disadvantage of ML models is the necessity of (sufficient) data, which are not always available. Rule models do not have this drawback. Finally, it has been shown that the appropriate hybridization of methods in a model leads to better results. In this dissertation, models and methods based on the above research directions will be presented. More specifically, the first two parts will present a framework, called ACRES, designed to generate rule-based expert systems from datasets, with a built-in model of certainty factors. ACRES utilizes several methods of extracting rules and calculating certainty factors from data, as well as organizing rules hierarchically to facilitate explanations. Its performance is comparable to many known ML methods. In the next part, a hybrid reasoning architecture is introduced, that combines several reasoning methods: rules, cases, weights, and voting, used to solve the complex problem of fish disease diagnosis, for which no pre-existing data were available. Finally, a hybrid reasoning method is presented, that integrates process mining and rule-based reasoning within a case-based reasoning framework. This method was applied to a personalized care system for individuals with autism spectrum disorders.
Description
Keywords
Expert systems, Rule extraction, Certainty factors, Uncertainty representation, Supervised learning, Hybrid reasoning, Intelligent information systems, Fish disease diagnosis, Autism spectrum disorders