Using Ensemble Learning to Improve Classification Accuracy in Medical Data

Abstract: Currently, electronic medical instruments are widely used in hospitals, medical polyclinics and doctors' offices to gather vital information about patients' bodies. Experts interpret medical data to distinguish the causes of illnesses. EEG is an example of a form of medical information that has many features. If the number of samples of patients is enlarged the volume of EEG data can increase dramatically and consequently exceed the limited capacity that can possibly be classified.  In order to solve the problems posed by the limitations of the current classification ability, SVMs are used. In some applications such as cognitive science, the accuracy rate of SVM classifiers is low. This fact is due to the complexity of the problem. The low accuracy rate may be caused by inappropriate feature space or the inability of classifiers to generalize results. SVM Ensembles can vastly improve generalization as, although some classifiers are not trained well enough to excel globally, they can at least achieve an acceptable local performance. This study's intention was to investigate the enhancement of classifier performance possible by applying SVM ensembles to classify two groups of data that were gathered during a type of healing operation known as Reiki, performed by a professional, and a placebo with an ordinary person pretending to perform it. Genetic algorithm is also the applied to this data to find the best features and feature combinations that reduce training time whilst increasing the correction classification rate.