A study of a group healthcare datasets in data mining domain

Abstract

This research is to search for alternatives to the resolution of complex medical diagnosis where human knowledge should be apprehended in a general fashion. Successful application examples show that human diagnostic capabilities are significantly worse than the neural diagnostic system. The study presents the particular case of analysis of eleven datasets containing data associated to several Healthcare datasets. The datasets are analyzed in various Healthcare domains to target different Medical areas. Paradigm of artificial neural networks is shortly introduced and the main problems of medical data base and the basic approaches for training and testing a network by medical data are described. There are eight algorithms used in this study, which are DT, SVM, RBF, MLP, k-NN, Naïve Bayes, Bayes Net and Logistic Regression. These eight algorithms have been performed with using 10-fold cross validation and train/test split over the eleven datasets. It’s also examined what is the effect of Principal Component Analysis inside the research. The performance metrics that are focused in this thesis are Percent Correct, True Positive Rate, False Positive Rate, Precision, Recall, F-Measure, AUC and Error Rates. As this is a benchmarking study for different classifiers and datasets, a special benchmarking criterion has been created for the evaluation of the thesis.