Model of Extreme Gradient Boosting Based Term Frequency (TFXGBoost) for Classifying Public Complaint Reports

Abstract

Various algorithms and machine learning techniques are being applied to improve the efficiency and effectiveness of the process of automatically classifying complaint reports from the public in Indonesia. One machine learning algorithm that has recently gained benchmarks in the state of the art of various problems in machine learning is eXtreme Gradient Boosting (XGBoost). This study aims to develop an extreme gradient boosting model based on term frequency (TFXGBoost) to predict whether a text is classified as a complaint or not a complaint based on the data studied. Based on the experimental results, TFXGBoost achieved 92.79% accuracy with eta / learning rate hyperparameters of 0.5, gamma of 0, and max_depth of 3 and the computation time required to adjust the hyperparameters was 13870.012468 seconds.