Comparison of Decision Tree J48 and CART in Liver Cancer Symptom with CARNEGIE-MELLON UNIVERSITY Data

Liver cancer remnants one of the leading cause of cancer-related deaths in United States and the world. There are many types of liver diseases, including various types of hepatitis, chronic liver disease, liver cirrhosis and liver cancer. Among them, hepatitis is the main cause of liver cancer. Therefore, we hope to understand the relationship between hepatitis and symptoms through data exploration. This report will be based on 155 patient data provided by CARNEGIE-MELLON UNIVERSITY in 1988, and will use the supervised machine learning model of classification and relationship rules to base actual cases on 20 different attributes of symptoms. Speculate whether a person died of liver disease. This study compares the two series of algorithms of J48 (Gain Ratio) and CART (Classification and Regression Tree) evolved from ID3 (Iterative Dichotomiser 3) in the classification tree in the decision tree with Gini index in the Java environment, and the data is pre-processed with normalization. By comparing all samples, cross-validation and 66% training data, J48 is better than CART in the average of the three comparisons with close to 87% accuracy rate, and CART has the highest correct rate in all samples with 90.3232% accuracy rate. Finally, it is found that there is no difference in deleting the attribute of the relevance of the Apriori algorithm. This research provides that doctors and scientists can use simple machine learning tools to obtain accurate results and prescribe medicines to the symptoms.