Information and Communication Engineering

Piata Binte Nahar

Thesis Title: Machine learning–based prediction model for oxygen requirement in hospitalized COVID-19 patients

Overview: Enter ProjeThe COVID-19 pandemic has created severe global shortages of essential resources, particularly oxygen, which is critical for saving lives in patients with respiratory complications. This study develops predictive models to forecast the need for oxygen therapy among hospitalized COVID-19 patients, aiming to ensure timely and adequate oxygen supply within healthcare facilities. We utilized real-world data from 1000 PCR-positive patients admitted to three hospitals in Dhaka, Bangladesh. The dataset included demographics, symptoms, laboratory results, and comorbidities. Six machine learning (ML) models were implemented, including logistic regression (LR), eXtreme gradient boosting (XGBoost), random forest (RF), decision tree (DT), K-nearest neighbors (KNN), and Naïve Bayes (NB). After data preprocessing and tenfold cross-validation, model performance was evaluated on a separate test set. Key predictors identified were oxygen saturation (SpO₂), respiratory distress, blood clots, pneumonia, hypertension, cardiovascular disease, kidney disease, lung infection, and age. Among the models tested, random forest demonstrated the best performance, achieving an accuracy of 93.69%, a recall of 97.5%, a precision of 93.9%, an F1-score of 97.8% and an AUC of 0.919. These results underscore the potential of machine learning techniques to enhance healthcare planning and resource management. By enabling early identification of oxygen needs, predictive models can reduce preventable deaths and strengthen healthcare resilience in future pandemics.