XGBoost 기반의 조기 중지를 활용한 광고 클릭 예측 방안

초록

Continuous data training on websites and social media platforms with machine learning algorithms that predict if a particular user clicks on ads results in better performance for training datasets, while test datasets experience overfitting problems that do not improve after a fixed number of learning iterations. In this paper, we propose an early stop of the learning process based on the XGBoost algorithm rather than the existing algorithm to avoid overfitting. XGBoost is a method to avoid overfitting by training complex data models, monitoring the performance of the models learned in a separate cluster of test data and stopping the training procedure if the performance of the test dataset has not improved after a fixed number of training iterations. We automatically select inflection points where the performance of the test dataset begins to decrease, thus implementing accuracy while avoiding overfitting, which continues to improve the performance of the training dataset according to the model's overfitting. Finally, the experimental results showed performance improvements based on the XGBoot algorithm compared to the Logistic Regression algorithm and the Decision Tree algorithm.

키워드