Hyperparameter Optimization Supported by SHAP Analysis for Performance Enhancement of Machine Learning Models in Covid-19 Diagnosis

The emergence of SARS-CoV-2 has led to increased scientific focus on developing effective diagnostic tools. Accurate detection is crucial for controlling the outbreak, and artificial intelligence (AI)-based methods have shown promise. This study uses machine learning (ML) techniques to predict COVID-19 from blood values, specifically, hemogram test results obtained from Van Yuzuncu Yil University Dursun Odabas Medical Center. Various ML algorithms were tested, with the Random Forest method achieving the highest accuracy. Model performance was further improved through optimization, where the Genetic Algorithm (GA) proved most effective. SHAP analysis was employed to enhance the interpretability of the predictions by identifying key features influencing the model’s decisions. Among the three evaluated datasets, Dataset 3 achieved the highest accuracy (91.56%). Dataset 2, after optimization, reached 85.09% accuracy with balanced performance, while Dataset 1 saw improved accuracy (65.02%) but lower recall. The GA-optimized model reached an AUC of 0.9467, indicating strong classification capability. These findings highlight the effectiveness of AI-driven models in disease detection and their potential to support healthcare systems by enabling faster and more accurate diagnosis. Future efforts will focus on integrating different modeling strategies and deep learning techniques to further improve diagnostic accuracy.