Performance evaluation of selected decision tree algorithms for COVID-19 diagnosis using routine clinical data

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Epidemiology

Background:

• The novel coronavirus disease (COVID-19) presents significant global public health and economic challenges.
• Early detection of COVID-19 is crucial for effective treatment and controlling viral transmission.
• Machine learning (ML) offers potential for enhancing COVID-19 identification.

Purpose of the Study:

• To compare seven decision tree (DT) algorithms for identifying the optimal clinical diagnostic model for COVID-19.
• To evaluate the performance of different DT models using established metrics.

Main Methods:

• A retrospective hospital-based dataset was utilized to train and evaluate seven distinct decision tree (DT) algorithms.
• Model performance was assessed using accuracy, sensitivity, specificity, Receiver Operating Characteristic (ROC) curves, and Precision-Recall Curves (PRC).

Main Results:

• The Gini Index identified key diagnostic criteria, including lung lesion existence, fever, and contact history.
• The J-48 decision tree algorithm achieved the highest performance metrics: accuracy=0.85, F-Score=0.85, ROC=0.926, and PRC=0.93.
• Lung lesion existence, fever, and contact history were identified as the most significant indicators for COVID-19 diagnosis.

Conclusions:

• The J-48 algorithm shows promise for clinical implementation in healthcare settings.
• Utilizing the J-48 model can enhance both the accuracy and speed of COVID-19 diagnosis.
• This study highlights the potential of machine learning in improving diagnostic capabilities for infectious diseases.