Applying Support Vector Machine to Electronic Health Records for Cancer Classification

An electronic health record (EHR) is a digital record of patients’ clinic information and medical records. The EHRs have been receiving more attentions, due to the extensive amount of valuable medical data, which is able to help medical providers and researchers diagnose and analyze the diseases and healthcare. Cancer diagnosis is very critical to potential patients since finding and treating cancers at an early stage can improve the survival rates. In this study, we use the EHRs obtained in clinics in New York City to classify the cancers. We train the support vector machine (SVM) model with medical records extracted from the EHRs for cancer classification. The model trained with 100 pieces of medical records per cancer, can achieve the prediction accuracy as 86.2% for 10 different types of cancers, and 97.33% for 3 types of cancers if trained with 400 pieces of medical records per cancer.