Bayesian belief networks for dementia diagnosis and other applications: a comparison of hand-crafting and construction using a novel data driven technique
Year: 2008
Type: dissertation
Abstract: The Bayesian network (BN) formalism is a powerful representation for \nencoding domains characterised by uncertainty. However, before it \ncan be used it must first be constructed, which is a major challenge \nfor any real-life problem. There are two broad approaches, namely \nthe hand-crafted approach, which relies on a human expert, and the \ndata-driven approach, which relies on data. The former approach is \nuseful, however issues such as human bias can introduce errors into \nthe model. We have conducted a literature review of the \nexpert-driven approach, and we have cherry-picked a number of common \nmethods, and engineered a framework to assist non-BN experts with \nexpert-driven construction of BNs. The latter construction approach \nuses algorithms to construct the model from a data set. However, \nconstruction from data is provably NP-hard. \nTo solve this problem, approximate, heuristic algorithms have been \nproposed; in particular, algorithms that assume an order between the \nnodes, therefore reducing the search space. However, traditionally, \nthis approach relies on an expert providing the order among the \nvariables \n--- an expert may not always be available, or may be unable to \nprovide the order. Nevertheless, if a good order is available, these \norder-based algorithms have demonstrated good performance. More \nrecent approaches attempt to ``learn'' a good order then use the \norder-based algorithm to discover the structure. To eliminate the \nneed for order information during construction, we propose a search \nin the entire space of Bayesian network structures --- we present a \nnovel approach for carrying out this task, and we demonstrate its \nperformance against existing algorithms that search in the entire \nspace and the space of orders. \nFinally, we employ the hand-crafting framework to construct models \nfor the task of diagnosis in a ``real-life'' medical domain, \ndementia diagnosis. We collect real dementia data from clinical \npractice, and we apply the data-driven algorithms developed to \nassess the concordance between the reference models developed by \nhand and the models derived from real clinical data.
Author: Lloyd Oteniya
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W604586230