magistrsko delo
Dejan Ognjenović (Author), Marko Robnik Šikonja (Mentor)

Abstract

Interakcije zdravil so prepletanja učinkov zdravil, ki lahko povzročijo želene ali škodljive učinke na pacientu. V nalogi smo z orodjem strojnega učenja iskali interakcije zdravil, ki lahko na zdravje bolnikov vplivajo negativno. Naš pristop k reševanju problema temelji na dveh algoritmih strojnega učenja. Pri tem smo upoštevali hierarhiji zdravil in bolezni ter bazo LexiComp. Algoritem posplošenih povezovalnih pravil poskuša s pomočjo hierarhij poiskati pravila, ki poleg osnovnih elementov upoštevajo tudi njihove posplošitve v hierarhiji. Drugi uporabljeni algoritem je iskanje pravil s koristnostno funkcijo, ki uporablja statistične informacije podatkov. Algoritme smo testirali na umetno generiranih podatkih in na realnih podatkih pacientov iz Univerzitetnega kliničnega centra v Ljubljani. Najdena pravila so pregledali farmacevti, ki so jih podrobno analizirali in komentirali. Rezultati algoritmov so obetavni, saj smo odkrili nekaj zanimivih novih pravil in vzorcev.

Keywords

strojno učenje;podatkovno rudarjenje;povezovalna pravila;interakcije;interakcije zdravil;zdravila;hierarhije;LexiComp;generalizacija;ontologije;posplošena povezovalna pravila;povezovalna pravila s funkcijo koristnosti;funkcije koristnosti;hiperkaliemija;računalništvo;računalništvo in informatika;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [D. Ognjenović]
UDC: 004.8:615.015.2(043.2)
COBISS: 1536161731 Link will open in a new window
Views: 777
Downloads: 229
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Searching for unwanted drug interactions
Secondary abstract: Drug interactions are interweaving effects between two or more drugs that can have desirable or harmful effects on patients health. In this thesis we for searched harmful drug interactions. Our approach is based on two machine learning algorithms for association rule mining. We use two given hierarchies, one for drugs (ATC), the other for diseases (ICD), and one proprietary interaction database LexiComp. A generalized association rule algorithm tries to find rules that contain basic elements as well as elements from given hierarchies. The second algorithm uses high-utility pattern mining. The utility function was designed to use statistical information from both the data and the hierarchies. Algorithms were tested on artificial data and on a dataset from University Medical Centre Ljubljana. Detected rules were reviewed, analyzed, commented and evaluated by pharmacists. The results are promising as several interesting new rules and patterns are detected.
Secondary keywords: machine learning;data mining;association rules;interactions;drug interactions;drugs;hierarchy;LexiComp;generalization;ontologies;generalized association rules;high utility association rules;utility function;hypercalemia;context;computer science;computer and information science;master's degree;
File type: application/pdf
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 63 str.
ID: 8739572