Biomedical subjects

S Dzeroski

Publications and source records attributed to S Dzeroski.

9 recordsLinked to original sources

PubMed2006

Radon in a thermal spring: identification of anomalies related to seismic activity.

Anomalies have been observed in the radon content of thermal spring water at the Italian-Slovenian border. To distinguish the anomalies caused by environmental parameters (air and water temperature, barometric and hydrostatic pressure, rainfall) from those ascribed solely to earthquakes with M(L) from 1.2 to 2.5 and epicentres, R(E), within 2R(D) (R(D)--Dobrovolsky's radius), two approaches have been used: (i) correlation between time gradients of radon concentration and hydrostatic pressure, and (ii) regression trees within machine learning programs. The regression trees approach has been improved by introducing additional environmental parameters and prolonging the measuring period.

Disasters↗

PubMed2003

Application of decision trees to the analysis of soil radon data for earthquake prediction.

Different regression methods have been used to predict radon concentration in soil gas on the basis of environmental data, i.e. barometric pressure, soil temperature, air temperature and rainfall. Analyses of the radon data from three stations in the Krsko basin, Slovenia, have shown that model trees outperform other regression methods. A model has been built which predicts radon concentration with a correlation of 0.8, provided it is influenced only by the environmental parameters. In periods with seismic activity this correlation is much lower. This decrease in predictive accuracy appears 1-7 days before earthquakes with local magnitude 0.8-3.3.

Decision Trees↗

PubMed2001

A reappraisal of saprobic values and indicator weights based on Slovenian river quality data.

The saprobic values and indicator weights used in the Slovenian saprobic system are reappraised using data from the 1990 to 95 river quality surveys of Slovenia. The conceptual basis of the reappraisal is described and then formulated mathematically. The analysis is based on 1,106 biological samples and covers 300 taxa. The results are expressed in terms of revised saprobic values and indicator weights that mirror the ones previously assigned by ecological experts. The most significant differences between original and revised values are highlighted and discussed. It is concluded that: (a) the revised values and weights are more representative of their 'true' values than are the original values and weights, but that it would be premature to consider them definitive; (b) the analytical method provides a sound data-based approach to the revision of saprobic values and indicator weights; and (c) the method could help to improve and harmonise the various saprobic systems currently in use across Europe.

Calibration↗

PubMed2001

Supporting discovery in medicine by association rule mining in Medline and UMLS.

The paper presents an interactive discovery support system for the field of medicine. The intended users of the system are medical researchers. The goal of the system is: for a given starting concept of interest, discover new, potentially meaningful relations with other concepts that have not been published in the medical literature before. The known relations between the medical concepts come from the Medline bibliographic database and the UMLS. We use association rules for discovering the relationship between medical concepts. We evaluated the system by testing how successfully it predicted future discoveries (new relations between concepts). We first divided the Medline database into two segments (older and newer) using the publication date. Then we calculated how many of the new relations found by the system in the older segment become known relations in the newer segment. We found out with statistical significance that the system predicts new relations better then someone predicting randomly. The evaluation showed that our approach for supporting discovery in medicine is successful, but also that some improvements are needed, especially on limiting the number of potential discoveries the system generates.

Algorithms↗

PubMed2000

Using data mining and OLAP to discover patterns in a database of patients with Y-chromosome deletions.

The paper presents a database of published Y chromosome deletions and the results of analyzing the database with data mining techniques. The database describes 382 patients for which 177 different markers were tested: 364 of the 382 patients had deletions. Two data mining techniques, clustering and decision tree induction were used. Clustering was used to group patients according to the overall presence/absence of deletions at the tested markers. Decision trees and On-Line-Analytical-Processing (OLAP) were used to inspect the resulting clustering and look for correlations between deletion patterns, populations and the clinical picture of infertility. The results of the analysis indicate that there are correlations between deletion patterns and patient populations, as well as clinical phenotype severity.

Chromosome Deletion↗

PubMed2000

A data mining approach to the development of a diagnostic test for male infertility.

The paper presents a database of published Y chromosome deletions and the results of analyzing the database with data mining and other heuristic techniques with the goal of developing a diagnostic test for male infertility. The database describes 382 patients for which 177 markers were tested. Two data mining techniques, clustering and decision tree induction were used, as well as a heuristic set cover algorithm. Clustering was used to group markers according to their appearance across patients, while a heuristic set covering algorithm was used to select as small a set of markers that cover as many patients with deletions as possible. This algorithm created a diagnostic set of 13 markers that cover more than 90% of the patients with deletions. Finally, decision tree induction was used to relate deletion patterns to the severity of the clinical phenotype. A decision tree induced from the data uses 5 markers, all of which are also in the diagnostic set of 13 markers, to show relations between the severity of the clinical phenotype and deletion patterns which have not been known previously.

Algorithms↗

PubMed1999

WWW search engine for Slovenian and English medical documents.

The information tool for the organization and searching of Slovenian and English medical documents is presented. The tool, partly still in development phases, performs automatic subject description of documents, searching with natural language queries and rankig of search hits according to their relevance. The search engine allows the searcher to use relevance feedback in order to perform incremental improvement of search results. The machine learning system TILDE for learning user profiles was also applied. Documents marked by the user as relevant or non-relevant are used to find characteristics that distinguish relevant documents from non-relevant ones.

Abstracting and Indexing↗

PubMed1996

Rule induction and instance-based learning applied in medical diagnosis.

Machine learning methods have been applied in a variety of medical domains in order to improve medical decision making. Improved medical diagnosis and prognosis can be achieved through automatic analysis of patient data stored in medical records, i.e., by learning from past experience. Given patient records with corresponding diagnoses, machine learning methods are able to classify new cases either through constructing explicit rules that generalize the training cases (e.g., rule induction) or by storing (some of) the training cases for reference (instance-based learning). This paper presents the methodologies of rule induction and instance-based learning and their application to medical diagnosis, in particular, the problem of early diagnosis of rheumatic diseases. It also discusses the possibility to use existing expert knowledge to support the learning process and the utility of such knowledge.

Algorithms↗

PubMedDate not supplied

Acquiring background knowledge for machine learning using function decomposition: a case study in rheumatology.

Domain or background knowledge is often needed in order to solve difficult problems of learning medical diagnostic rules. Earlier experiments have demonstrated the utility of background knowledge when learning rules for early diagnosis of rheumatic diseases. A particular form of background knowledge comprising typical co-occurrences of several groups of attributes was provided by a medical expert. This paper explores the possibility of automating the process of acquiring background knowledge of this kind and studies the utility of such methods in the problem domain of rheumatic diseases. A method based on function decomposition is proposed that identifies typical co-occurrences for a given set of attributes. The method is evaluated by comparing the typical co-occurrences it identifies as well as their contribution to the performance of machine learning algorithms, to the ones provided by a medical expert.

Algorithms↗