Covid-19 Readmission Prediction using Different Feature Selection Techniques and Machine Learning Models

In this paper, a retrospective study was carried out with real-world patient data from Universiti Malaya Medical Centre (UMMC), Kuala Lumpur, Malaysia. After pre-processing, a final total of 441 patient data remained, with 399 non-readmitted cases, and 42 readmitted cases. The date of admission ranged from February 2021 to October 2021. The readmitted cases made up 9.52% of the total data collected. Two types of feature selection techniques were applied, namely Recursive Feature Elimination (RFE) and Chi Square Test. The common features of both feature selections are BMI, heart rate, systolic blood pressure, age, respiration rate, diastolic blood pressure, and SPO2. Selected features were fed into 2 machine learning models for model training, including logistic regression and Support Vector Machine (SVM). Comparing the feature selections, Chi Square Test outperformed RFE in both logistic regression and SVM. As the best result, when using features selected using Chi Square Test, SVM achieved the highest accuracy of 0.981, with AVC of 0. 942.