COVID-19 from symptoms to prediction: A statistical and machine learning approach

Area of Science:

• Data Science
• Epidemiology
• Medical Informatics

Background:

• The COVID-19 pandemic highlighted the need for data-driven public health strategies.
• Analyzing patient data is crucial for understanding disease patterns and informing interventions.

Purpose of the Study:

• To predict COVID-19 patient age groups using statistical and machine learning techniques.
• To identify associations between patient symptoms and age demographics.
• To evaluate the effectiveness of various machine learning and ensemble methods for prediction.

Main Methods:

• Utilized a dataset of over 10,000 anonymized COVID-19 patient records.
• Applied statistical tests (ANOVA, t-tests) for variable assessment.
• Employed machine learning models: Decision Tree, Naïve Bayes, KNN, Gradient Boosted Trees, SVM, Random Forest.
• Implemented ensemble methods: bagging, boosting, and stacking.
• Performed rigorous data preprocessing to optimize model performance.

Main Results:

• Identified significant associations between key COVID-19 symptoms and patient age groups.
• Ensemble methods, especially stacking with Random Forest as a meta-learner, substantially improved prediction accuracy (0.7054).
• Stacking enhanced K-Nearest Neighbors (0.529 to 0.63) and Naïve Bayes (0.554 to 0.622) performance.

Conclusions:

• Machine learning, particularly ensemble stacking, offers a powerful approach for predicting COVID-19 patient age groups based on symptoms.
• Findings can inform targeted public health strategies, resource allocation, and treatment protocols for different age demographics.
• Integrating predictive models into clinical settings supports real-time responses and interventions during pandemics.