Application and Comparison of Artificial Neural Networks and XGBoost on Alzheimer's Disease

Abstract

Alzheimer's disease (AD) is a kind of brain disease, which causes abnormal memory loss, thought chaos, and behavior confusion. There are still no effective methods or medicine to prevent the worsening of AD. The best way at present is to reduce the risk of getting AD. In this paper, the author constructs an artificial neural network (ANN) and XGBoost to determine whether or not a person gets AD, by analyzing how related factors impact the group a person belonging to. The Open Access Series of Imaging Studies (OASIS) longitudinal MRI data were analyzed cross age, gender, education, social economic status (SES), mini mental state examination (MMSE), estimated total intracranial volume (eTIV), clinical dementia rating (CDR), normalized whole brain volume (nBWV), and atlas scaling factor (ASF). The purpose of this study is to decide whether a person is demented or not by comparing two classic methods, thus to explore the advantages and disadvantages of two models in real world application. The analysis is helpful to predict and model the different features in non-demented and demented people, therefore giving a clearer perspective for reducing people's risk of dementia by making appropriate adjustments. The accuracy of testing with ANN is 89.3%, with 37 matched non-demented and 30 matched demented out of 75 observations, which is 20% testing set. The accuracy of fitting the data is 93.3% for XGBoost, with 38 matched non-demented and 32 matched demented out of 75 observations. K-fold cross validation is applied to improve the accuracy rate. The accuracy is then improved to 95.6% for ANN and 99.6% for XGBoost. In conclusion, the result is consistent with the former literature study, showing that the machine learning method is more accurate than deep learning.