Differential Diagnosis between Degenerative and Vascular Dementia using Data Mining Algorithms

Abstract

In recent years, the population of Taiwan is aging rapidly. As per Ministry of Interior data (2008), the older population has amounted to 2.4 million accounting for 10.4% of the total population. Dementia is one of the most common disorders among the elderly and there are different types (degenerative, vascular etc.) of dementia; therefore, accurate and early diagnosis of dementia with differentiation of “dementia type” is crucial for treatment. Data mining algorithms like Logistic Regression, Classification and Regression Tree (CART) and Support Vector Machine (SVM) can be used to differentiate demented patients in to degenerative and vascular type of dementia. We collected 533 samples. Out of these samples, 420 were used to train the model and others for testing data. The accuracy of models was 0.6991(Logistic Regression), 0.6903(CART) and 0.7345(SVM) based on variables like gender, orientation, registration, language and drawing. Further analysis and testing was accomplished with four additional variables, namely, age, education, attention and short term memory which resulted with accuracy of 0.6372, 0.7168 and 0.9027 for Logistic Regression, CART and SVM respectively. From these results, we can conclude that: 1. SVM is better than Logistic regression and CART in both cases 2. High dimensions algorithm like SVM gives better result than low dimensions algorithms (Logistic regression and CART).