Enhanced Contrast Pattern Based Classifier for Handling Class Imbalance in Heterogeneous Multidomain Datasets of Alzheimer Disease Detection

Abstract

Alzheimer’s disease is a kind of dementia which progresses deterioration in the intellectual abilities and systematic actions of elderly persons. Discovering Alzheimer at its initial stage may help the victims to improve their quality of life so that its degeneration could be delayed. There are many existing classification models used to detect Alzheimer, but the accuracy of detection rate is not promising because handling medical data often face challenges when it has the dataset with class imbalance. Without considering the class imbalance in the classifier is used it performs the prediction bias for the majority class lead to adverse consequences. Hence, this paper focuses on overcoming class imbalance in AD dataset by devising Enhanced Contrast Pattern Based Classifier (ECPBC) which provides equal importance to the instances with majority class and minority class. This proposed model represents a strong contrast knowledge by finding the interesting measure and pruning redundant instances for developing accurate and robust classifier. This work used two different datasets one is collected from sensor data of smart home known as CASAS and another one is OASIS dataset which has MRI information of patients. Each attributes importance is determined by applying mutual information the irrelevant attributes are eliminated and only significant attributes are used for classification. The simulation results proved that by handing the class imbalance very prominently the proposed ECPBC model produced more accuracy on Alzheimer disease detection compared to other standard classifiers.