| Abstract: |  The aim of this thesis is to evaluate performance and usability of selected Artificial Immune Systems (AIS) on the problem of classification and automatic processing of loosely structured free-text medical records. After a review on the state-of-the-art AIS algorithms, we have selected and implemented Artificial Immune Recognition System (AIRS) and Negative Selection Algorithm (NSA) algorithms as representatives. After preliminary testing and behaviour study we have altered the algorithms to fit provided datasets using (among others) modified distance metric based on the Damerau-Levenshtein distance. On the datasets sized 22 000 and 1 500 000 words, we have obtained the following best classification accuracy: 78.17 %, 65.80 % respectively for the AIRS and 81.22 %, 64.49 % respectively for the NSA.
|
|---|