Comparative Analysis of Classifiers for the Assessment of Respiratory Disorders Using Speech Parameters

Non-invasive techniques for the assessment of respiratory disorders have gained increased importance in recent years due to the complexity of conventional methods. In the assessment of respiratory disorders, machine learning may play a very essential role. Respiratory disorders lead to variation in the production of speech as both go hand in hand. Thus, speech analysis can be a useful means for the pre-diagnosis of respiratory disorders. This article aims to develop a machine learning approach to differentiate healthy speech from speech corresponding to different respiratory disorders (affected). Thus, in the present work, a set of 15 relevant and efficient features were extracted from acquired data, and classification was done using different classifiers for healthy and affected speech. To assess the performance of different classifiers, accuracy, specificity (Sp), sensitivity (Se), and area under the receiver operating characteristic curve (AUC) was used by applying both multi-fold cross-validation methods (5-fold and 10-fold) and the holdout method. Out of the studied classifiers, decision tree, support vector machine (SVM), and k-nearest neighbor (KNN) were found more appropriate in providing correct assessment clinically while considering 15 features as well as three significant features (Se > 89%, Sp > 89%, AUC > 82%, and accuracy > 99%). The conclusion was that the proposed classifiers may provide an aid in the simple assessment of respiratory disorders utilising speech parameters with high efficiency. In the future, the proposed approach can be evaluated for the detection of specific respiratory disorders such as asthma, COPD, etc.