Abstract
Parkinson's disease is a neurological syndrome that manifests slowly and gradually, making it difficult to diagnose at an early stage. Voice alterations can be used as a detectable marker of early detection. The Synthetic Minority Oversampling Technique (SMOTE) is employed to address class imbalance issues in the datasets. For optimal feature selection, a novel approach called Fisher Score-Based Recursive Feature Elimination (FRFE) is proposed, and it is compared with state of art feature selection methods namely correlation coefficient, mutual information, backward feature elimination, and recursive feature elimination. The performance of models was evaluated across different classifiers using two voice datasets, with different features so as to confirm that FRFE works for any dataset irrespective of features. The FRFE performs better than the state of methods of comparison in terms of accuracy and variance.
TopLiterature Survey
Many researchers have worked on the Parkinson’s disease classification problem. Few researchers have used only base classifiers to predict the disease and few have used hybrid ensemble models to classify the disease, which showed higher performance measures. A literature survey is done to consider all factors that impact the disease to classify it correctly and efficiently.
Ishu et al. (2022) proposed an approach for predicting Parkinson's disease using a random forest classifier, combined with Principal Component Analysis (PCA) for dimensionality reduction. The accuracy of the proposed PCA-RF model was compared with an ANN model that also utilized PCA. The paper discusses the benefits of using dimensionality reduction to remove redundancy in data and reduce the time and storage needed for analysis. The proposed approach achieved a significant accuracy of up to 90%, and the performance of the model was found to depend on the importance of features and the dimension of the dataset. Overall, the paper presents a promising approach for predicting Parkinson's disease that could be useful in clinical settings.
Nilashi et al. (2022) present a comparative study of machine learning techniques for predicting Motor-UPDRS and Total-UPDRS scores in Parkinson's disease (PD) patients. The methods used include data preprocessing, dimensionality reduction using PCA, clustering using ensembles of EM, and prediction using ensembles of SVR. The paper highlights the importance of categorizing PD progression into different stages using Movement Disorder Society-Unified Parkinson's Disease Rating Scale (MDS-UPDRS) scores, which range from 1 to 5, and the Hoehn & Yahr Scale, which is a widely used measure of disease severity. The paper provides insight into the effectiveness of different machine learning techniques for predicting UPDRS scores and could be useful in clinical settings for monitoring PD progression.
Saravanan et al. (2022) present a detailed review of various machine learning and deep learning-based AI techniques applied to Parkinson's disease (PD) diagnosis and their impact on opening up new research avenues. The study explores data-driven AI technologies' opportunities and current status in PD diagnosis. The methods used in the paper include diagnosis based on various modalities such as gait, EEG, spiral drawing, neuroimaging, and physiological signals using machine learning and deep learning techniques. The paper highlights the high accuracy achieved in the diagnosis of PD based on physiological signals using machine learning techniques. However, the paper also points out the limitations of certain physiological signals like EEG and speech, which possess low spatial resolution and are susceptible to contamination by artifacts. The paper emphasizes the need for selecting appropriate metrics for evaluating ML model’s performance in PD classification and exploring different metrics. Overall, the paper provides valuable insights into the potential of AI-based techniques in PD diagnosis and highlights the challenges and opportunities for future research.
Key Terms in this Chapter
Variance Score: Variance Score is the proposed performance measure. A high variance score indicates that the built model is highly accurate and reliable.
RFE: Recursive Feature Elimination which is the Wrapper technique. This technique determines how the main elements in the dataset explain variation by fitting a model. Each iteration removes features that aren't as essential one by one after establishing their relevance.
RE: Redundancy Elimination which is the technique of iteratively picking the most relevant attributes from the parameters of a learned ML model.
SMOTE: Synthetic Minority Oversampling Method is a branch of statistics for balancing data by equally augmenting a lot of instances in it.
LightGBM: Light Gradient-Boosting Machine, is a distributed gradient-boosting framework for machine learning that is free and open-source. It is used for classification, ranking, and other machine learning challenges and is based on decision trees.
FRFE: Fisher score-based recursive feature elimination (Fisher score-based RFE) is the proposed feature selection method that is used to select the most relevant feature for model building.
Parkinson's Disease: Tremors, rigidity, and difficulty walking, balancing, and coordination are all symptoms of Parkinson's disease.
Fisher Score: Fisher Score is a statistical measure that is used to select the most relevant features.