Analyzing the effect of data preprocessing techniques using machine learning algorithms on the diagnosis of COVID-19

Area of Science:

• Medical diagnostics
• Machine learning in healthcare
• Data science

Background:

• Real-time polymerase chain reaction (RT-PCR) tests are insufficient for rapid COVID-19 diagnosis due to global spread.
• Medical data, including COVID-19 blood parameters, presents challenges like inconsistency, incompleteness, and large scale.
• Limited and potentially biased medical datasets can adversely affect machine learning study accuracy.

Purpose of the Study:

• To investigate the impact of various data preprocessing techniques on the classification accuracy of COVID-19 using blood parameters.
• To improve existing, limited datasets of COVID-19 blood data for more consistent machine learning results.
• To compare the effectiveness of different imputation and balancing methods in enhancing COVID-19 classification models.

Main Methods:

• Applied feature encoding and scaling to a dataset of 279 patients' blood data.
• Utilized K-nearest neighbor (KNN) imputation and Multiple Imputation by Chained Equations (MICE) to handle missing data.
• Employed Synthetic Minority Over-sampling Technique (SMOTE) for data balancing.
• Evaluated the performance of ensemble (bagging, AdaBoost, random forest) and popular classifier algorithms (KNN, SVM, logistic regression, ANN, decision tree).

Main Results:

• The highest accuracy achieved with the bagging classifier was 83.91% using KNN imputation without SMOTE.
• With SMOTE, the bagging classifier achieved accuracies of 83.42% (KNN imputation) and 83.74% (MICE imputation).
• Comparative analysis demonstrated that specific data preprocessing combinations significantly influence classification success.

Conclusions:

• Data preprocessing techniques critically impact the accuracy of machine learning models for COVID-19 classification from blood data.
• The choice of imputation (KNN vs. MICE) and balancing (SMOTE) methods affects model performance.
• Experimental results highlight the importance of selecting the right combination of preprocessing steps for successful COVID-19 diagnostic models.