Abstract:

Abstract An increasing number of people have died as a result of the COVID-19 pandemic's second wave of breakout. As has been shown, several nations' healthcare systems are being destroyed by the second wave. Regional routine testing combined contact tracing can take the place of regional constraints in preventing the virus from propagating, and the "Track, Test, and Treat" programme has straightened the epidemic track in its early phases. Thus, to lower infection rates and minimise the negative effects on medical Machine learning along with feature engineering is a potential domain for developing Covid 19 positive as well as negative samples classification, a critical research objective in contemporary engineering. While there are effective machine learning-based methods to classify COVID-19 positive and negative samples like cough audio signals, detection accuracy with the highest possible sensitivity and specificity is still not scalable using the majority of contemporary methods. Typically, detection accuracy is proportional to the optimal features used to train the classifier. As a result, it is obvious that optimizing features for Covid 19 infection recognition from cough audio signals is a possible research objective. In support of this argument, this article suggested and described a novel technique “Adaptive Boosting based Supervised Learning (ABSL) Approach for Covid-19 Prediction from Cough Audio Signals”. The spectral features and Mel-frequency cepstral coefficients are used in the proposed model. The feature engineering has been done by the diversity assessment model “kruuskal-wallis test”. In addition, a novel binary classification strategy has been derived by using adaptive boosting strategy.   The experiments have been done on benchmark dataset to evaluate the proposed approach's performance against a comparable contemporary method Random forest classifier that trained by Mel-frequency cepstral coefficients (MFCCs). The experiments demonstrated that the suggested ABSL has the potential to escalate prediction accuracy with a low rate of false alarms.

Keywords: