Project 21: Data mining applied to data on respiratory effects of air pollution Organization: Leiden University and the National Institute Abstract of Public Health and the Environment (RIVM), the Netherlands Aim of the projects is the investigation and Primary Contact: Maarten Lamers development of techniques for dealing with specific Address: Leiden University aspects of health-related data in epidemiological Dept. of Computer Sciences studies. Non-linear neural network data-mining Niels Bohrweg 1 methodologies were successfully devised for 2333CA Leiden ¢ finding heterogeneity in response between The Netherlands unknown subpopulafions in a study; Emall: maarten@wi.leidenuniv.ul ¢ reducing non-linear correlations between related URL: http://www.wi.leidenuniv.nl/CS/ and bio-medical variables; http://www.fivm.nl/sector2/ccm/index.html ¢ testing for non-linearity of responses in multi-level population studies. Duration: 4 years Applications to data from studies on short-term Number of People: 3 respiratory effects of air-pollution have illustrated the Tools Developed: prototyped neural network techniques merits of these methods when complementing Academic Disciplines: Computer Science, Neural traditional methods. Computation, and (Environmental) Epidemiology Keywords: children, lung function, air pollution, neural networks Project Related Publications: M.H. Lamers, and J.N. Kok, and E. Lebret, "Combined Neural Network Models for Epidemiological Data: Modelling Heterogeneity and Reduction of Input Correlation", Proceedings of the International ConJbrence on Artificial Neural Networks and Genetic Algorithms ~TCANNGA'97), G.D. Smith, and N.C. Steele, and R.F. Albrecht (Editors), 147-151, Springer-Verlag, 1998. M.H. Lamers, and J.N. Kok, and E. Lebret," A Multilevel Nonlinearity Study Design", Proceedings of the 1EEE International Joint Con~rence on Neural Networks (IJCNN'98) Anchorage, AK, 730-734 1998. Project 22. Multistrategic data mining in a clinical meningoencephalifis database Organization: Japanese AI Society, KBS Research Group Contact: Shusaku Tsumoto Address: Medical Research Institute, Tokyo Medical and Dental University 1-5-45 Yushima, Bunkyo-ward Tokyo 113 Japan Email: tsumoto@computer.org URL: http://www.kde] jnfo.eng.osaka-cu.ac.ip/SIGKB S/ Duration: 0.5 years Number of People: 10 Tools: Developed: SIBLE, KDD-R, ProbRough Purchased: C4.5, SAS Academic Disciplines: Medical Informatics Abstract This project aims at inducing as much information as possible from clinical databases. Until now, the following six methods are applied to a clinical database (meningoencephalitis): 1. association rule induction, 2. decision trees with genetic algorithm based feature selection, 3. decision tree induction with random sampling, 4. rough set based rule induction, 5. logit analysis 6. linear discriminant analysis. Experimental results show that knowledge which medical experts did not expect was extracted by different methods. The research will be extended to other methods and databases. Keywords: discretization, rule induction, clinical databases, mulfistrategic data mining

/lp/association-for-computing-machinery/multistrategic-data-mining-in-a-clinical-meningoencephalitis-database-WgTXEl3ndX