Combining knowledge from different sources

Abstract

The present paper deals with the problem of an assessment of symptoms in medical diagnosis. A unified interpretation of symptoms is necessary to estimate their significance in a diagnosis. Yet, even if they are properly defined, different evaluations of them based on experts' knowledge or statistical estimation are possible. The present study aims at combining evaluations that may originate from an expert or can be found from statistical features of the data, as well as those determined for 'easy' and 'difficult' diagnostic cases. A model of diagnostic inference is proposed in the framework of the Dempster-Shafer theory extended for fuzzy focal elements. The basic probability assignment defined in this theory estimates weights of symptoms. Two basic probability assignments can be created and then combined. In this way weights of symptoms represent knowledge common for two kinds of data or obtained from an expert and from data. Thus, a combination of heuristics and data mining results becomes possible. An algorithm of the basic probability assignment calculation is suggested and tested for medical data: a database from the internet and individually gathered data.

Membership Functions for Fuzzy Focal Elements

“…The algorithm described in [28] makes it possible to determine thresholds η BPA and η T to ensure the best accuracy of a diagnosis. The algorithm with some changes can be also used to proceed iterative elimination of rules from a full rule set.…”

Computer recognition systems

“…Straszecka (2010), in her paper ‘Combining knowledge from different sources’, deals with the problem of an assessment of symptoms in medical diagnosis. A unified interpretation of symptoms is often necessary to estimate their significance in a diagnosis.…”

Identification of I‐equivalent subnetworks in Bayesian networks to incorporate experts' knowledge

“…Recent studies that use expert knowledge have addressed this problem to improve the structure learning capability of BNs (Büyüközkan, Kayakutlu, & Karakadılar, ; Jun & Kim, ; Straszecka, ; Werhli & Husmeier, ). These studies have sought to combine the probabilistic outcomes from a BN and knowledge from domain experts to gain improved representation of their causal relationships (Heckerman, Geiger, & Chickering, ).…”