A comparison of SMOTE-based machine learning algorithms for predicting hospital mortality in traumatic patients
Code: G-1092
Authors: Maryam Zamani ℗, Mahnaz Yadollahi *, Vahid Ahmadipanah
Schedule: Not Scheduled!
Tag: Biomedical Signal Processing
Download: Download Poster
Abstract:
Abstract
Introduction: Trauma is one of the most critical public health issues worldwide, leading to death and disability and affecting the age groups. Thus there‘s great interest in models to prognosticate mortality in trauma patients. The present study aims to explore the potential of developing and evaluating SMOTE-based machine-learning tools for predicting hospital mortality in trauma patients with imbalanced data. Methods: This cross-sectional study included 475 cases referred to Shahid Rajaee Hospital, selected through census sampling, between October 2022 and June 2023. The data were extracted from the patient's medical records. According to the unbalanced nature of the data, SMOTE approaches were used. further Decision Tree(DT), Random Forest(RF), K- Nearest Neighborhood( KNN), Canonical Neural Network( CNN), Support Vector Machine( SVM), Extreme Gradient Boosting(XGBoost) and Logistic Regression( LR) methods were used to predict in- hospital mortality of patients with traumatic injuries. Metrics similar to accuracy, precision, recall, and area under the receiver operating characteristic curve( AUC) were used to estimate the performance of the algorithms. Results: Most patients were male( 446,93.89), their mean age was32.26 ±0.21 years, and The mean follow-up time from the date of trauma to the date of outcome was4.14 ±5.21 days. The performance of ML algorithms isn't good with imbalanced data, whereas the performance of SMOTE-based ML algorithms is significantly improved. The results showed that XGBoost performed the best among the prediction approaches. While the synthetic minority oversampling technique( SMOTE), a type of over-sampling, also demonstrated a certain position of performance, under-sampling was superior. Overall, predicting by the XGBoost model with samples using SMOTE had the best results. Conclusion: This study presented the findings of an empirical comparison of sampling techniques and classification algorithms that can impact the accuracy of imbalanced samples by combining two techniques. ML algorithms based on SMOTE may be more effective than other ML algorithms in predicting the outcome of traumatic injuries.
Keywords
Imbalanced Data, Machine Learning Algorithms, SMOTE
