Lung Cancer Classification Using Random Oversampling and Gradient Boosted Decision Tree

Abstract

Lung cancer has the highest number of sufferers in men, especially in Indonesia. An unhealthy lifestyle, smoking, and pollution also aggravate the patient's condition. In this study, a diagnosis was made of patients with suspected lung cancer. For an experiment, the data from public datasets, “Cancer Patient," “Survey Lung Cancer,” and “Cancer_Data.” The research phase includes exploratory data analysis (EDA), pre-processing, and classification. EDA aims to know data types, missing values, correlations between attributes, and outliers. Pre-processing consists of data cleaning and data discretization. In the next process, we use randomized oversampling to overcome imbalanced data. The final step was classification using Gradient Boosted Decision Tree (GBDT). The experiment scenario uses imbalanced and balanced data. For the testing scenario, the variation in learning rate and the number of trees were used with Randomized Search Tuning. The distribution of training and testing data uses 5-fold cross-validation. The result shows that using balanced data between classes is better than imbalanced data. In addition, we also classify the dataset with the k-nearest neighbor and support vector machine. The GBDT produces better performance for two datasets.