An Adaptive Cancer Prognosis Framework for Cholangiocarcinoma based on Machine Learning Techniques

From the observation in 2014, cancer was the number one cause of deaths in Thailand and has been continuously increased until present. Cholangiocarcinoma is the subset of liver cancer, which is one of the top five cancers founded in Thailand. To reduce the risk of cholangiocarcinoma in patients, we need to find the factors of causing cancer and predict the probability of cancer earlier. However, the probability of cholangiocarcinoma is less than 1%. This causes imbalance classification problem in machine learning. In this paper, we propose an adaptive cancer prognosis framework for cholangiocarcinoma based on machine learning techniques, namely “CanWiser”. CanWiser is used to automatically learn the patient dataset (e.g., demographics and laboratory test results), pre-process data, oversample data to solve the imbalance problem using SMOTE, generate the prediction models using classification of machine learning techniques (i.e., support vector machine, decision tree, naïve Bayes, and random forest), and predict the probability of cholangiocarcinoma. The proposed framework can generate the prediction model providing the sensitivity 75%, specificity 83.41%, and accuracy 83.34%. CanWiser also provides the personalized recommendation for patients to reduce the risk of cholangiocarcinoma. Moreover, our proposed framework can adaptively learn and generate the models, which can fit for the new dataset.