A Comparative Study of Machine Learning Algorithms for Optimizing Ventilator Pressure Prediction in ICU Patients

During the COVID-19 pandemic, many patients had severe respiratory distress and required ventilator assistance to survive. One of the most widely used treatments in the ICU to assist patients' breathing is the use of ventilators. Ventilators are machine that pump air into and out of a patient's lungs. However, the use of ventilators in ICU patients must be done carefully because it can cause several complications. This study utilized machine learning techniques to predict the necessary ventilator pressure for patients in intensive care units. In this study, the effectiveness of three different algorithms—linear regression, CatBoost, and XGBoost—was evaluated. These algorithms were chosen due to their diverse strengths: linear regression for its simplicity and interpretability, CatBoost for its handling of categorical variables, and XGBoost for its gradient boosting framework. The most accurate method was XGBoost, which had a significantly lower mean absolute error (MAE) than the other two algorithms. XGBoost's MAE for predicting the test set is 1.0425, significantly lower than CatBoost's (1.2107) and Linear Regression's (4.717). XGBoost as well as comparable algorithms can be used to more effectively control ventilator support, which will ultimately enhance patient outcomes and give guidance for critical care decision-making.