Ensemble learning model for diagnosing COVID-19 from routine blood tests.

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Computational Biology

Background:

• The COVID-19 pandemic necessitates rapid and reliable diagnostic tools.
• Routine blood tests offer potential for early COVID-19 screening.
• Existing diagnostic methods require improvement for widespread application.

Purpose of the Study:

• To develop and validate an ensemble learning model for COVID-19 diagnosis using routine blood tests.
• To enhance the accuracy and efficiency of early COVID-19 screening.
• To provide a robust tool for timely patient care and disease containment.

Main Methods:

• An ensemble learning model (ERLX) was developed, combining Extra Trees, Random Forest, and Logistic Regression classifiers with an XGBoost meta-classifier.
• Data preprocessing involved KNNImputer for missing values, Isolation Forest for outlier detection, and SMOTE for class balancing.
• SHAP values were utilized for model interpretability and feature importance analysis.

Main Results:

• The ERLX model achieved exceptional performance on a public dataset, with 99.88% overall accuracy and 99.38% AUC.
• The model demonstrated high sensitivity (98.72%) and specificity (99.99%), crucial for reliable screening.
• ERLX significantly outperformed existing state-of-the-art models on the same features.

Conclusions:

• The ERLX model offers a robust and highly accurate solution for early COVID-19 screening using routine blood tests.
• Its superior performance metrics suggest potential for clinical deployment in rapid patient assessment.
• The model contributes to combating the COVID-19 pandemic through improved diagnostic capabilities.