Classification performance analysis of intelligent systems for biomedical data

One of the most popular areas of medical informatics is computer assisted analysis of biomedical data. Expert systems that are so called computer based disease diagnosis systems support medicians in disase diagnosis decision making. Fastening the medical decison phase with preserving accuracy is possible with the expert systems that are trained with medicians knowledge and experience. In the literature, there is an aboundant of related work that are suitable for the task of supportive expert systems.Intelligent systems constitute an important part of expert medical decision systems. An expert medical decision system?s accuracy depends on the performance of the intelligent system that is the kernel of the software. Therefore, the accuracy of the expert system is one to one correspondent with the performance of the intelligent system. Hence, it is important to determine the factors that affect the performance of the intelligent systems while analyzing medical data, in order to develop high accurate medical decision systems.This thesis is fulfilled with the aim of determining performance factors of about thirty intteligent system algorithms while analyzing ovarian cancer, prostate cancer, Parkinson disease, dermatology and diabet datasets. We selected feature selection, data pre-processing, algorithm parameter changes and ensemble learning out of so many performance effecting factors for medical data analysis. As a result of many experiments carried out, a correlation is found with the mentioned factors and the performances of the intelligent systems. In this way, it is proved computationaly that the performance of the experts systems to be developed as medical decision systems migth be improved with feature selection, data pre-processing, algorithm parameter change and ensemble learning.In this work, the experimented intelligent system algoritms and feature selection strategies are realized using Matlab, Weka and Microsoft Visual Studio software development environments.