A Predictive Modelling Approach in the Diagnosis of Parkinson's Disease Using Cerebrospinal Fluid Biomarkers

Abstract

The research in Parkinson's disease {PD) using biomarkers has long been dominated by measuring dopamine metabolites or alpha-Synuclein in cerebrospinal fluid. However, these markers do not allow early detection or monitoring of disease progression. In the recent years, metabolic profiling of body fluids has become powerful and promising tools in identification of the novel biomarkers in the diagnosis of the disease. While not much research has been done using machine learning techniques and predictive modeling to predict the severity of Parkinson's disease. The purpose of this project is to apply a predictive modeling approach in the diagnosis of Parkinson's disease using Cerebrospinal Fluid Biomarkers. The dataset for this study was collected from the PPMI website which comprises of 360 - Parkinson's patient, 220 - Control and 20 - SWEDD (Scans without evidence for dopaminergic deficit). Various predictive models were developed in order to classify the disease based on its severity. The various machine learning algorithms used in this process are Decision tree, Random forest, Support Vector Machine {SVM), K- Nearest Neighbor (KNN), and Gradient boosting. Feature scaling and Mean normalization was applied to standardize the dataset. The above mentioned machine learning algorithms were applied on the Parkinson's Progression Markers Initiative (PPMI) data and accuracy for each algorithm was calculated. Out of all the models, Random forest and Gradient Boosting gave the best classification accuracy of 66.67%. In conclusion, the main factors that might have affected accuracy of the model are dataset size, missing data and number of features. To sum up, while the results show some predictive power, we conclude negative results and hence these models are not Clinically significant.