Lymphedema Detection Based on Self-Reported Symptoms for Breast Cancer Survivors

With the increased breast cancer survival, more women are facing life-time risk of developing lymphedema. Using web-based technology to develop low cost and pragmatic patientcentered e-platform will enable lymphedema detection using realtime symptom report and greatly improve patient-centered care. We have conducted a machine learning study using a dataset consisting of 25 symptom features and lymphedema status of 355 patients, aiming to classify patients into with and without lymphedema using the symptom features only. We implemented and compared 5 well-known supervised learning algorithms including Decision Tree variants (C4.5, C5.0), Gradient Boosting Model, Artificial Neural Network and Support Vector Machine. Furthermore, we have conducted feature selection on our dataset based on various methods. Several symptom importance rankings are provided for better understanding of association between the symptoms and the lymphedema status.