A Survival Prediction Model of Self-Immolation Based on Machine Learning Techniques

Background: Self-immolation is one of the violent methods of suicide in developing countries. Predicting the survival of self-immolation patients helps develop therapeutic strategies. Today, machine learning is widely used in diagnosing diseases and predicting the survival of patients. This study aims to provide a model to predict the survival of self-immolation patients using machine learning techniques. Materials and Methods: A retrospective cross-sectional study was conducted on 445 hospitalized self-immolated patients admitted to a burn hospital between March 2008 and 2019. Python programming language version 3.7 was used for this goal. All possible machine-learning algorithms were used. Gradient Boosting, support vector machine (SVM), random forest, multilayer perceptron (MLP), and k-nearest neighbors algorithm (KNN) were selected as the high-performance machine learning technique for survival prediction, and then they were compared by evaluation metrics such as F1 score, accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC). Based on this comparison, the best model was reported. Results: SVM was the best algorithm. F1 score, accuracy, and AUC for this machine-learning model were 91.8%, 91.9%, and 0.96, respectively. The machine learning model results revealed that surgical procedures, score, length of stay, anatomical region, and gender obtained the most important and had more impact than other factors on patients' survival prediction. Conclusion: In this paper, machine learning algorithms were used to create a model for survival of self-immolation patients. The results of this study can be used as a model for predicting self-immolation patients' survival, better treatment management, and setting up policies and medical decision-making in burn centers.