Comparison of machine learning algorithms to increase prediction accuracy of COPD domain

Abstract

Medicine and especially chronic diseases, like everything else on earth is filled with ambiguity. This is why, identifying patients at risk present a big challenge to human brain. Poor control and misdiagnosis of chronic diseases has a great impact quality of life of patients, the expenses and performance of health care system. The global economic cost of chronic diseases could reach {dollar}47 trillion by 2030, according to a study by the World Economic Forum (WEF). Beside this economic burden, such treatment failure increases the risk of progression of disease which inevitably leads to premature death or further illness and suffering. Today, health informatics is reshaping the research in the medical domain due to its potential to concurrently overcome the challenges encountered in the traditional healthcare systems. Uncertainty, accuracy, causal attributes and their relationship, all have their places in this new technology through contemporary machine learning algorithms. Prediction of exacerbation of Chronic Obstructive Pulmonary Disease (COPD) is considered one of the most difficult problems in the medical field. In this paper, we will leverage unused machine learning methods to increase prediction accuracy in COPD. To this end, we compared three of the most common machine learning algorithms (decision tree, naive Bayes and Bayesian network) based on ROC metric. Furthermore, we used discretization process for the first time in this context.