Parkinson's Disease Subtype Classification: Application of Decision Tree, Logistic Regression and Logit Leaf Model

摘要

Parkinson's Disease has two subtypes which are Tremor Dominant (TD) and Postural Instability/Gait Difficulty (PIGD). Each subtype has the difference in clinical treatment, so it is necessary to classify Parkinson's Disease subtypes. Three classification methods were implemented: decision tree, logistic regression, and logit leaf model (LLM). Data on 229 people with early Parkinson's disease from the PPMI (Parkinson's Progression Markers Initiative) database were used in the analysis. Imbalanced data problem were handled using oversampling, undersampling, SMOTE (Synthetic Minority Over-sampling Technique). Logistic regression with SMOTE using parameter set-up α = 600, γ = 200 produced the best result, according to the accuracy of 98.3 %, sensitivity of 98.41 %, and specificity of 99.07 %.