Evaluation of predictive learners for cancer incidence and mortality

Ability to project cancer incidences and mortality is very important for cancer research and healthcare planning and is also a vital part of cancer screening and management programs. In current years, machine learning algorithms have been successfully shown to generate high forecasting accuracy and have drawn interest from healthcare professionals, research community, planners and policy makers. In this paper, three supervised machine learning classification techniques are compared to project cancer incidence and mortality rates. Classification methods covered in this work are Bayesian Network, Naïve Bayes and K-Nearest Neighbor. These classification methods have been tested on the datasets provided by Statistics Canada. This paper evaluates the performance of above classification techniques that examine the accuracy of each method via the utilization of prediction accuracy measures including Mean Absolute Errors, Root Mean Squared Errors, Relative Absolute Error, Precision, ROC area, TP and FP Rate.