<![CDATA[Abstract. Purpose: The purpose of this study is to improve accuracy on the prediction of default risk. Defaults on p2p lending platforms result in significant losses for lenders and pose a threat to the efficiency of the overall peer-to-peer lending system. It is then very important to have an understanding of the methods capable of managing such risks Even with only a slight improvement in the accuracy of the model's ability to anticipate default significant losses can be avoided.Methods: The method used to make predictions is a combination method of stacking ensemble models with the LightGBM metalearner as the final estimator. While the base-learner algorithms used are Multi Layer Perceptron (MLP), Support Vector Machine (SVM) and Random Forest (RF).Result: The PPDai dataset used in this study is P2P lending data from the PPDai platform from China. The data split process uses the 10-fold cross validation method. Evaluate the model using a confusion matrix that generates accuracy values. The evaluation results showed that Stacking-LightGBM as the best model in the PPDai dataset received an accuracy of 87.18%.Novelty: This research shows that the accuracy of the peer-to-peer lending default prediction model can be improved using the Stacking-LightGBM method. The stacking ensemble method can beat the accuracy of a single classification algorithm in terms of making predictions. The use of meta-learners can improve the performance of ensemble stacking models.]]>
Copyrights © 2023
