On defining rules for cancer data fabrication

Abstract

Data is essential for machine learning projects, and data accuracy is crucial for being able to trust the results obtained from the associated machine learning models. Previously, we have developed machine learning models for predicting the treatment outcome for breast cancer patients that have undergone chemotherapy, and developed a monitoring system for their treatment timeline showing interactively the options and associated predictions. Available cancer datasets, such as the one used earlier, are often too small to obtain significant results, and make it difficult to explore ways to improve the predictive capability of the models further. In this paper, we explore an alternative to enhance our datasets through synthetic data generation. From our original dataset, we extract rules to generate fabricated data that capture the different characteristics inherent in the dataset. Additional rules can be used to capture general medical knowledge. We show how to formulate rules for our cancer treatment data, and use the IBM solver to obtain a corresponding synthetic dataset. We discuss challenges for future work.