Accurate Classification of COVID-19 Based on Incomplete Heterogeneous Data using a KNN Variant Algorithm.

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Classifying coronavirus disease 2019 (COVID-19) is crucial amid the ongoing pandemic.
• Medical datasets are often incomplete and heterogeneous, posing challenges for traditional classification algorithms.
• Existing K-Nearest Neighbors (KNN) algorithm modifications address data issues but have limitations.

Purpose of the Study:

• To introduce a novel K-Nearest Neighbors variant (KNNV) algorithm for accurate COVID-19 classification.
• To leverage rough set theory for handling data incompleteness and heterogeneity.
• To identify an optimal value for K using rough set techniques.

Main Methods:

• Developed a KNNV algorithm utilizing rough set theory to manage incomplete and heterogeneous medical data.
• Employed Euclidean and Mahalanobis distance metrics for enhanced classification scope.
• Tested the KNNV algorithm on a real-world COVID-19 patient dataset from the Italian Society of Medical and Interventional Radiology.

Main Results:

• The KNNV algorithm demonstrated efficient and accurate classification of COVID-19 cases.
• Experimental results showed superior performance compared to three other KNN derivatives.
• KNNV significantly outperformed competitors in precision, recall, accuracy, and F-Score.

Conclusions:

• The proposed KNNV algorithm offers a robust solution for classifying COVID-19 from complex medical datasets.
• The methodology is adaptable for classifying other diseases and general classification tasks beyond the medical field.