Advancing Machine Learning in COVID-19 Diagnostics: Symptom-Based Classification and Ensemble Techniques

Abstract

The COVID-19 pandemic necessitated the development of diagnostic methods that are not only rapid and accurate but also capable of distinguishing COVID-19 from similar respiratory diseases. The use of ensemble learning has proven effective in enhancing diagnostic accuracy through detailed symptom analysis. Previous studies have relied on traditional machine learning techniques like logistic regression and decision trees for early detection. These methods often struggle with symptom overlap, a challenge that ensemble learning addresses by combining predictions from various models to improve diagnostic precision. This study implements an ensemble learning framework that integrates diverse models to refine the accuracy of COVID-19 diagnoses based on patients' symptoms. The process includes data preprocessing, feature engineering, and optimizing ensemble methods such as random forests and gradient boosting. The Symptoms Based COVID-19 Classification Algorithm offers a straightforward diagnostic approach by assessing symptom proportions against a threshold. Conversely, the COVID-19 Detection Using an Ensemble Learning Model employs a sophisticated ensemble of models, enhancing diagnosis through weighted symptom analysis. The Symptoms Based Algorithm achieves 85% accuracy with some limitations in specificity, whereas the Ensemble Learning Model shows superior performance, achieving 90% accuracy and effectively minimizing false positives. Although the Symptoms Based Algorithm is useful for quick assessments, the Ensemble Learning Model's accuracy and comprehensive analysis make it more suitable for clinical application. Future efforts will focus on integrating broader data sources and validating these models in practical scenarios.