IMPLEMENTATION OF ENSEMBLE MACHINE LEARNING MODEL FOR POSTOPERATION COMPLICATIONS PREDICTION

Abstract

Background. The paper raises the problem of predicting postoperative complications. Complications after surgery can lead to deterioration of the patient's condition, increased treatment costs and even death. Therefore, being able to predict possible complications after surgery helps in making decisions about choosing the best treatment approach and providing more effective postoperative care. Materials and methods. To predict complications after surgery, machine learning models were studied, including logistic regression, the decision tree method, the random forest method, the k-nearest neighbor method, the support vector method, a multilayer perceptron with a pre-selected architecture and the weighted average voting ensemble method. Results. Preprocessing of depersonalized data on patients, informative indicators were selected and an ensemble machine learning model based on the random forest method, the support vector method, a multilayer perceptron with a pre-selected architecture was trained. Conclusions. As a result of the ensemble method (weighted average voting – hard voting), the accuracy increased to 78,8 %.

Keywords

• machine learning
• forecasting
• postoperative complications
• random forest