Dementia Prediction Model Based on Feature Selection, SMOTE and Data Mining Algorithm

107

Dementia is a heterogeneous syndrome characterized by common symptoms such as cognitive impairment and functional disability, it’s caused by a progressive degeneration of brain cells and occurs in middle-aged and elderly people. There will be a lesion in patient’s brain. The brain function is affected enough to interfere with the person''s normal social or working life. There have been many scholars using various machine learning algorithms, such as Artificial Neural Network (ANN), Support Vector Machine(SVM), Random Forest in researches on dementia, hoping to improve the accuracy of neuro psychology testing result via machine learning method and establish dementia predictive classification model via dementia classification sample. Establishing predictive classification model helps to achieve the goal of early detection and early treatment. Once there is sample of new dementia classification sample, the established predictive classification model can be used to process sample classification and medical treatment and provides timely medical service. The raw data was provided by the medical institutes. The data set is the test results of Cognitive Abilities Screening Instrument (CASI). The disease types were classified and merged. Data discretization and Synthetic minority over-sampling techniques were adopted to pre-processing collected data. To find the important features, the study used five feature selection techniques: Information Gain, Gain Ratio, Chi-Square Test, Relief-F, and OneR. Artificial Neural Network (ANN), Polynomial Support Vector Machine, C4.5 classification, Linear Support Vector Machine and Logistic Model Trees (LMT) are machine learning algorithms which were adopted to establish the predictive model and validate the accuracy of Dementia data set.