Automatic COVID-19 Prediction with Comprehensible Machine Learning Models

Abstract

Background The development of technology has a significant impact on every aspect of life, whether it is the medical industry or any other profession. The potential of artificial intelligence has been demonstrated in data processing and analysis is used to inform decisions in the field of health care. The most crucial action is the early detection of a life-threatening illness to stop its development and spread. Highly contagious COVID-19 is a disease that requires immediate attention as it has spread globally. There is a need for a technology that can be utilised to detect the virus because of how quickly it spreads. With the increased use of technology, we now have access to a wealth of COVID-19-related information that may be used to learn crucial details about the virus. Objective The objective of the work is to develop comprehensible machine learning models for the automatic prediction of COVID-19. These models aim to accurately predict the likelihood of COVID-19 infection based on relevant input features, such as symptoms, demographics, and potential diagnostic tests or imaging results. Methods In this work, we mainly aimed to assess how well different machine learning methods might predict COVID-19 situations. In order to do this, we thoroughly evaluated a variety of widely used classifiers in machine learning. Popular algorithms like the random forest, k-nearest neighbour, and logistic regression were all included in our analysis. Results To assess the performance of our suggested algorithms using different machine learning techniques, we used an open-source dataset in the study. Our algorithms performed better than other models that are currently in use, which is noteworthy. The high degree of precision in predicting COVID-19 instances is demonstrated by our remarkable accuracy of 96.34%. We also obtained a good F1 score of 0.98 for our models, indicating the strength and efficiency of our method in obtaining metrics for both recall and precision. Conclusion This work highlights the possibility of understanding machine learning algorithms for COVID-19 prediction automatically. We have shown that techniques such as logistic regression, random forest, and k-nearest neighbor methods may reliably predict COVID-19 situations while preserving interpretability. In order to promote acceptance and confidence among healthcare professionals and enable well-informed decision-making in clinical settings, these models' transparency is essential. To improve these models' efficacy and scalability in the future, more research will be needed to enhance and validate them on a variety of datasets. In the end, utilizing understandable machine learning algorithms presents encouraging opportunities for COVID-19 early diagnosis and control, supporting international public health campaigns.