Modified mutual information feature selection algorithm to predict COVID-19 using clinical data

Area of Science:

• Biomedical Informatics
• Computational Biology
• Infectious Disease Research

Background:

• The COVID-19 pandemic highlighted the critical need for rapid and accurate disease detection.
• Blood tests are essential diagnostic tools due to SARS-CoV-2's impact on hematological parameters.
• Effective machine learning models for COVID-19 prediction depend on selecting relevant diagnostic features.

Purpose of the Study:

• To develop an optimized feature selection method for COVID-19 prediction using blood test data.
• To enhance the accuracy and generalizability of machine learning-based diagnostic models.
• To identify the most informative features from blood test results for disease classification.

Main Methods:

• Proposed Modified Mutual Information (MMI) for feature relevance ranking and optimal subset selection.
• Employed a backtracking algorithm within MMI to refine feature selection.
• Utilized Support Vector Machines (SVM) for robust classification of COVID-19 cases.

Main Results:

• The MMI feature selection method combined with SVM achieved a high prediction accuracy of 95%.
• This approach demonstrated superior performance compared to other existing feature selection techniques.
• The model exhibited strong generalizability across diverse benchmark datasets, indicating reliable diagnostic potential.

Conclusions:

• Modified Mutual Information (MMI) is a highly effective technique for selecting relevant features in COVID-19 blood test analysis.
• The integration of MMI with Support Vector Machines (SVM) offers a powerful and accurate tool for disease prediction.
• This study provides a validated computational approach for improving diagnostic accuracy in pandemic scenarios.