COVID-19 Detection via Blood Tests using an Automated Machine Learning Tool (Auto-Sklearn)

Abstract

     Widespread COVID-19 infections have sparked global attempts to contain the virus and eradicate it. Most researchers utilize machine learning (ML) algorithms to predict this virus. However, researchers face challenges, such as selecting the appropriate parameters and the best algorithm to achieve an accurate prediction. Therefore, an expert data scientist is needed. To overcome the need for data scientists and because some researchers have limited professionalism in data analysis, this study concerns developing a COVID-19 detection system using automated ML (AutoML) tools to detect infected patients. A blood test dataset that has 111 variables and 5644 cases was used. The model is built with three experiments using Python's Auto-Sklearn tool. First, an analysis of the Auto-Sklearn process is done by studying the impact of several learning settings and parameters on the COVID-19 dataset using different classification methods, namely meta-learning, ensemble learning, and a combination of ensemble learning and meta-learning. The results show that using Auto-Sklearn with a meta-learning and ensemble learning parameter model predicts the patients infected with COVID-19 with high accuracy, reaching 96%. Furthermore, the best algorithm selected is the Random Forest Classifier (RF), which outperforms other classification methods. Finally, AutoML can assist those new to data sciences or programming skills in selecting the appropriate algorithm and hyperparameters and reducing the number of steps required to achieve the best results.