Prediction of COVID-19: WAOptimizer Ensemble Classification Model with Clinical Parameters

Abstract

Introduction:

Machine learning-based prediction systems have enormous potential to enhance clinical utility in diagnosing COVID-19. Machine learning can address prediction challenges in the healthcare sector by improving diagnostic efficiency and accuracy. Delivering high-quality services is challenging. Effective illness management and precise diagnosis are critical elements of healthcare.  Machine learning is transforming better progression toward prediction. The healthcare industry is adopting machine learning to improve efficiency. This technique is essential in healthcare to detect patterns within large datasets and diagnose the disease. Previous studies have predicted COVID-19 mortality using blood biomarkers and machine-learning approaches. The outcome of the prediction method effectively predicted the non-linear relationships among blood biomarkers. In addition, the prediction includes traditional assessment techniques for monitoring pulmonary diseases, such as X-rays and CT scans. Prompt detection and virus diagnosis are essential for infection control and reducing mortality rates.

Objectives:The primary goal of this research is to predict COVID-19 positivity and negativity based on blood test data by developing a stacking ensemble classifier algorithm WA-COVID Optimizer.

Methods:This research focuses on predicting the positive and negative statuses of the disease using a blood count dataset. To achieve better prediction performance, the study aims to develop an ML-driven diagnostic framework for early-stage COVID-19 diagnosis utilizing an ensemble stacking classification method. Several supervised machine learning methods are commonly used for predictions. These include Random Forest, LightGBM, Support Vector Machine, Logistic Regression with Lasso and Ridge regularization, XGBoost, AdaBoost, Gradient Boosting Machine, Multilayer Perceptron, Deep Neural Networks, and K-Nearest Neighbors. These models are combined to construct a stacking ensemble classification model that acts as a meta-model, leveraging the strengths of the base models.

Results:The performance metrics accuracy, precision, recall, F1-score, and the area under the ROC curve (AUC-ROC) are used to evaluate the prediction model. The Matthews correlation coefficient (MCC) assesses the ROC performance metrics. The proposed stacking ensemble classifier achieved an accuracy of 85%, an AUC-ROC of 90%, an MCC of 0.66, a precision of 81%, a recall of 85%, and an F1-score of 83%.

Conclusions:We developed a new data-driven strategy, the WA-COVID Optimizer, which synergizes multiple base models with a boosting mechanism. The proposed stacking classifier, WA-COVID Optimizer, predicted the best accuracy of 84% and a ROC AUC -90% for COVID-19-positive cases. The MCC validates the classifier performance, and the evaluation score is 66%.