Predictive models for colorectal cancer: A United Kingdom study
Oidamae, Ken Tobiko
The role ofMachine Learning (ML) and Artificial Intelligence (AI) in healthcare and other aspects of life is growing every day. The purpose ofthis study was to build predictive ML models to determine whether patients with colorectal cancer live or die and to draw insights on them. This study was of a causal design. Patients from the UK of various ages, different times of diagnosis and with different grades and stages of cancer were the focus of the study. These coupled with their vital status - alive or dead - were used to build the models. Three models were built - regression model, decision tree and extreme gradient boosting ensemble trees. The three models had various accuracies in predicting the survival with the regression model performing the worst followed by the decision tree and then the ensemble trees. Apart from the models, feature importance showed the significance of attributes like the stage of cancer and grade of tumor differentiation had on thelikelihood of a patient surviving or dying.
Submitted in partial fulfillment of the requirements for the Degree of Bachelor of Business Science in Actuarial Science at Strathmore University