Prediction analysis for Parkinson disease using multiple feature selection & classification methods.

Autor: Hema, M. S., Maheshprabhu, R., Reddy, K. Sudheer, Guptha, M. Nageswara, Pandimurugan, V.
Zdroj: Multimedia Tools & Applications; Nov2023, Vol. 82 Issue 27, p42995-43012, 18p
Abstrakt: Diagnosis of Parkinson's disease (PD) is a difficult undertaking that requires a variety of testing and clinical trials. Despite all these testing and trials, there is a considerable risk of misdiagnosis of Parkinson's disease, which causes a delay in making decisions about the best treatment for the patients. Computer aided diagnosis can improve the experiences and helps the doctors to predict PD patients. Four sorts of classification algorithms with various types of feature selection approaches are proposed in this paper for successful PD diagnosis. Support Vector Machine-SVM, Nave Bayes-NB, K-Nearest Neighbor-KNN, and Random Forest-RF are the categorization techniques employed. All the dataset's features are not necessary for classification. To choose the relevant characteristics, four feature selection approaches are used. Least Absolute Shrinkage and Selection Operator (LASSO), Backward-forward, rough set, and tree-based feature selection techniques are used. Experimentation is carried out using the Parkinson's Progressive Markers Initiative (PPMI) dataset. Four classification algorithms are constructed, together with and without feature selection procedures, and comparative research is conducted. To measure the performance of the classifiers, various evaluation methodologies are used. Overall Random Forest classifier with four feature selection methods' gives best results with average accuracy is 96% and rough set feature selection for all three classifiers' average accuracy is 97%. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index
Externí odkaz: