Voice disorders are characterised by voice quality and automatic voice quality detection driven by Machine Learning (ML) algorithms allows for non-invasive and inexpensive assessment of voice quality. The objective of this study is to identify the voice qualities of the given unlabeled sustained voice samples and the problem is modeled as a multi-class classification problem with four labels - Hoarse, Harsh, Breathy, and Healthy. The proposed method uses sustained voice samples of vowel /a/ from which acoustic features (Mel-frequency Cepstral coefficients (MFCCs), Spectral centroid, Spectral roll-off, ZCR, RMS, Chroma stft, Spectral contrast, and Melspectrogram) are extracted to train the ML classifiers (Bagging Classifier, ExtraTrees Classifier, GB, XGB, RF, LR, and DT) to classify the voice quality of the given unlabeled sustained phonation samples. The proposed models are evaluated on a newly created dataset consisting of sustained voice samples of vowel /a/ of different voice qualities collected from Kannada-speaking individuals. Among the proposed ML models, ExtraTrees Classifier trained on Spectral contrast and Melspectrogram features achieved an accuracy of 89.53% using the train-test set in the ratio 80:20.