Searching for Taxonomy-based Similarity Measures for Medical Data

Identifying an appropriate measure of similarity between concepts is an important step in solving data mining tasks and designing decision support systems, being also a challenging problem because of the plethora of measures in the current use. When choosing a similarity measure the particularities of data to be processed and the availability of additional information should be taken into account. This paper addresses the problem of estimating the (dis)similarity between lists containing entities defined by a taxonomy and presents the results of a comparative analysis of several measures. The analysis is conducted in the context of processing lists of ICD-10 diagnostic codes. The set of analyzed similarities is obtained by combining several code-level and listlevel measures. A weighted version of the measure based on edge counting is proposed and its ability to explain observed structures in data is estimated using some cluster validity indices. The comparative analysis involves also...

• Type: Journal
• Year: 2015
• Where: BCI
• Keywords: BCI 2015, Information Technology