Validation and Optimization of supervised machine learning model for rapid COVID-19 Diagnoses using clinical symptoms

Abstract

This research focuses on the development, validation and optimization of a supervised machine learning trained model for the rapid diagnosis of COVID-19 disease detection using patient’s clinical symptoms. Using the logistic regression algorithm, it is a linear model which is known for its robustness and accurate result, we aimed to improve the diagnostic process by improving its capability to handle missing data values from the dataset, assess feature importance, and mitigate overfitting and underfitting problems. The model was instructed and evaluated using a medical dataset from Kaggle, with four performance metrics including accuracy, precision, recall and F1 scores calculated. The results found using LR model are high accuracy (98.5%), with a perfect precision (100%) and a recall of approximately (96.9%), diagnosing the patients effectively by using the model in accurately identifying positive cases. ROC curve analysis revealed a high area under the curve (AUC), confirming the excellent differentiation among positive and negative cases. Also, the Precision-Recall curve illustrated the high precision and robustness of this model. Future work will focus on expanding the dataset, comparing with other different types of models, refining the hyperparameters, and validating the model in real clinical settings to further improve its diagnostic capabilities. This model represents a significant advance in the rapid and accurate detection of COVID-19 infectious disease, with the potential to improve diagnostic efficiency and support public health efforts.