Abstract:
This independent research brings together knowledge between data science and medicine. The two main objectives are 1.) to study the use of machine models to predict medical data, and 2.) to study the effectiveness of using algorithms to increase the amount of medical data without biasing the data to a specific point. Currently, factors predicting failure in non-surgical treatment of swallowing bowel syndrome have been studied to assess the risk and potential errors associated with treatment. The study of non-surgical gut swallowing found that there were more likely factors associated with predictive failure to treat non-surgical swallowing bowel syndrome, and there were some unique features of continuous numerical data that were difficult to calculate by hand. Machine learning models need to be studied to help predict failure in non-surgical treatment of swallowing bowel syndrome. The results of the study found that prediction with Non-Oversampling The best model was Logistic Regression, with an accuracy of 80% and a sensitivity of 86%. The model performs significantly better. The method of Oversampling that gives the model the best overall performance is K-means SMOTE, and the prediction model that gives the most accuracy is Decision Tree. When applying the machine learning model and the method of Oversampling, The model that gave the highest accuracy was Decision Tree with SMOTE-NC Oversampling and Decision Tree with K-means SMOTE Oversampling with an accuracy of 94%, and the model with the highest sensitivity was Support Vector Machine with Non-Oversampling that gives a sensitivity of 100%