NORMALISATION AND DIMENSIONALITY REDUCTION TECHNIQUES TO PREDICT PARKINSON DISEASE USING PPMI DATASETS

摘要

Parkinsonism is a syndrome consisting of motor complications such as resting tremor, bradykinesia, rigidity, posture difficulties and loss of postural reflexes. The disease is slowly caused by the loss of dopamine neurons in the substantial nigra and leads to issues with movement. The availability of large medical datasets provides researchers great opportunities to study the various diseases and provide predictions for better treatment. The proposed work uses PPMI dataset from Michael Fox, in which Unified Parkinson's disease Rating Scale (UPDRS) is most commonly used consistent scale to access Parkinsonism also UPDRS provides baseline assessment for the disease. In this paper, normalisation and dimensionality reduction are performed for prediction of Parkinson disease. The work proposed is to normalise and reduce dimensions of motor symptoms features of Movement Disorder Society-Unified Parkinson's disease Rating Scale (MDS-UPDRS) datasets provided by PPMI. The min-max normalisation is preferable than that of Z-score normalisation, since Z-score normalisation does not normalise exactly on the same scale. Dimension reductions for the various participants are performed against categories with PCA, NMF, Fast ICA and Kernel PCA. The Gaussian Mixture Model used for PPMI dataset provides best model using mixture of multi-dimensional Gaussian probability distribution.