Event Detail

Event Type: 
M.Sc. Presentation
Thursday, June 2, 2022 - 11:00 to 13:00
Kidder 274 and Zoom - contact nikki.sullivan@oregonstate.edu if you would like the link to attend via Zoom

Speaker Info

Local Speaker: 

Machine learning models are powerful tools which may aid in the prediction of survival outcomes of cancer patients. This study evaluated sixteen supervised machine learning models, eight classification and eight regression, on their ability to predict survival outcomes on breast cancer and prostate cancer data sets from the SEER database. The most accurate models, based on the fraction of correct predictions, were found to be the Gradient Boosting models for both classification and regression. To maximize accuracy, the hyperparameters of these two models were optimized. The computational time to train the models and make predictions was evaluated. Further improvements on the models, specifically the Linear Regression model, using SVD dimension reduction and low-rank approximations successfully decreased computational time. The results of this study support the development of future predictive models for cancer outcomes, particularly in the extension of the dimension reduction technique. Successful predictive models would benefit the tens of millions of cancer patients and medical professionals worldwide.