Predicting Breast Cancer Survivability Rates For data collected from Saudi Arabia Registries

The application of data mining and machine learning in directing clinical research into possible hidden knowledge is becoming greatly influencial in cancer research. This research presents a comparison of three data mining classification models: multi-layer perceptron neural networks, C4.5 decision trees and Naive Bayes. The classification models are built for breast cancer survivability prediction. The data set used is collected from registries across Saudi Arabia. Due to data scarcity, data synthesis had to be performed using a random seed and a double sampling procedure. After sufficiently preprocessing the data, the classification models were built and three performance measures were used to rank the models: Accuracy, Sensitivity and Specificity. The experiment was set up with multi-layer perceptron as the baseline scheme and with statistical significance of 0.05. Decision Trees performed marginally better than multi-layer perceptron and Naïve Bayes performed significantly worse than the baseline scheme. The result showed that Decsion tree is the the most accurate predictor for breast cancer survivbility in Saudi Arabia (Accuracy 0.979% ).