Voice Disorder Detection and Classification Using Machine Learning Techniques and Feature Selection Methods

Voice disorders could have a substantial effect on an individual's quality of life. The traditional methods of diagnosing voice disorders rely on subjective assessments by clinicians, which is time-consuming and prone to variability. To address this issue, we require an automated system which could accurately and objectively detect voice disorders using machine learning approaches. This research paper aims to design and implement a ML model that can analyses voice recordings and classify them as either normal or indicative of a voice disorder. The paper begins with a comprehensive literature survey highlighting the potential of ML and DL models in accurately detecting and classifying voice disorders. The paper then presents the planned workflow, which contains collection of data, preparation, feature extraction, data splitting, model selection and training, hyperparameter tuning, model evaluation, and voice disorder detection. The methodology section describes the dataset to be used, which includes a collection of voice recordings from individuals with voice disorders and healthy individuals. The implementation section introduces diverse ML classifiers, including SVM, Random Forests, k-NN, Naive Bayes, and decision trees. Feature selection methods, such as information gain, correlation-based feature selection, L1 regularization, and PCA, are also discussed. Finally, the paper outlines the results and discussions section, which explains the performance of the ML classifiers assessed with respect to accuracy, sensitivity, specificity, and the area under the ROC curve. The outcomes are discussed, highlighting the efficiency of the proposed machine learning model in detecting voice disorders.