Machine Learning-Based Prediction on Relapse of Acute Myeloid Leukemia
Yingzhe Li, Shiwei Gao, Zerong Guo, Liu Li, Youzhi Xiong, Sanshan Sun, Hui Li · 2024
-Acute myeloid leukemia (AML) is a high-incidence adult acute leukemia. Most patients will relapse after remission and turn into refractory and relapsed acute leukemia. This study aims at employing machine learning algorithms to develop a relapse prediction model for AML patients. Based on the AML patient data provided by the TARGET project from Meshinchi Laboratory of the National Institutes of Health (NIH) Cancer Research Center, we first normalize all samples to meet the requirements of the machine learning model. To eliminate the accuracy loss of the prediction model caused by category imbalance, we further use the synthetic minority over-sampling technique (SMOTE) to oversample the normalized data. Finally, we utilize the training datasets to develop three typical prediction models, namely Gaussian kernel support vector machine (G-SVM), linear kernel support vector machine (L-SVM), and decision tree (DT). The experiment results show that the data processed by SMOTE-based over-sampling can significantly improve the performances of the three prediction models, and the holistic performance of the G-SVM model is superior to another, which can be applied to the diagnosis and treatment decisions to assist clinicians in taking early intervention measures for AML patients.