Comparison of Predictive Models for the Prognosis of Lung Cancer

Abstract

The purpose of this research is to examine the efficiency of data mining techniques by creating a model to predict the likelihood of lung cancer. In the research procedure, the 310 rows 16 variables data was analyzed under the Cross Standard Process for Data Mining (CRISP-DM) and divided into two groups (teaching and testing data) using the 10-fold cross-validation method. Four data mining techniques consist of k-Nearest Neighbors, Decision Tree, Random Forest, and Vote Ensemble to evaluate algorithms for optimization value for each set of parameters in the model and compare the data classification performance by Accuracy, F-measure, Sensitivity, Specificity and Positive Predictive Value with RapidMiner Studio Version 10. The results show that technique of grouping method is the most suitable technique to build a model for trend prediction of cancer with the highest accuracy of 91.57%, which is very high level. The efficiency value is 60.61%, which is moderate level. The sensitivity is 51.67%, which is moderate level. And the specific value is 97.42%, which is very high level. The positive predictive is 73.08%, which is high level. Finally, the results of this research can be used to build a trend prediction for lung cancer patients. It is a preliminary screening of patient information before reaching the doctor.