Predicting liver disorder based on machine learning models

Abstract

As the main detoxification organ of human body, liver is very important in humans' health by metabolizing a lot of substances that are taken in, including alcohol and the medicine. However, if a person consumes too much alcohol or contaminated food, it will lead to liver disorder by causing little ingestion of essential nutrients. Accurate prediction for alcohol consumption, therefore, is very important by providing doctors the necessary information for diagnosing liver diseases. To address this problem, this paper introduces machine learning models to predict liver disorder. In addition, to alleviate the influence of data randomness by splitting the data set into a training set and a testing set, the leave-one-out cross valuation is utilized. The feature importance and the relationships between different features are also analyzed. The experimental results showed that the machine learning models are effective for alcohol consumption prediction. Among them, the random forest has the best performance in terms of accuracy (80.35%). The reason could be that the ensemble strategy used is helpful to reduce the over-fitting problem caused by the imbal-anced data set. This indicates that the random forest could be a useful tool for liver disorder prediction by providing helpful information or suggestions for doctors for diagnosing liver diseases.