ABSTRACT
Parkinson disease (PD) is diagnosed based on the number of clinical trials and tests. At present there exist no perfect methods for diagnosing PD. Sometimes, it may lead to misdiagnosis, and it causes delays in decision making and treatment. Computer-aided methods support doctors in diagnosing PD in a better way. In this chapter, a PD prediction and drug recommendation approach is proposed. It consists of three steps, namely feature selection, prediction, and clustering. Many attributes are available in the PD dataset. But, all attributes are not included for classification and prediction. Inappropriate attributes will lead to performance degradation and it improves the response time. Rough set and principal component analysis (PCA) are proposed to extract the required feature from the original dataset. The second step is the classification of PD patients’ data. A deep learning algorithm is proposed to predict the PD patients’ dataset. It will differentiate the case controls from the healthy individuals. The performance of the deep learning approach is compared with traditional supervised classification algorithms. The outcomes of the prediction, in addition to the regular test results conducted for the patient, will help the doctors to improve decision making. The last step is clustering the predicted Parkinson patients for their personalization. The k-means clustering algorithm is proposed to group the PD patients. The outcome of the clustering may be used for drug recommendations of the individual patients. The Parkinson's Progressive Markers Initiative (PPMI) dataset is applied for experimentation. The PD patients’ demographic information, motor, non-motor, and family history data is considered for analysis. The various performance metrics such as accuracy, specificity, and sensitivity are used to evaluate the performance of the proposed algorithms.
