IEEE-CIS Fraud Detection Based on XGB

As a result of the world switching to using credit cards in place of cash due to the quick advancement of technology, fraud incidents have increased. Fraud deals with circumstances in which there is criminal intent, yet it is usually difficult to discern. Much research indicate that global losses based on credit card fraud will exceed $35 billion by 2020. The credit card’s providers or those financial banks should protect users from any fraud risk they might confront. As a result, this work provides a machine learning-based strategy to identify fraudulent transactions using data from the Kaggle-obtained IEEE-CIS Fraud Detection dataset. The model combines three most efficient ensemble models including Categorical Boost (CatBoost), Extreme Boost (XGBoost) and LightGBM (LGBM). Instead of training the model directly, this paper provides detailed data preprocessing and feature engineering methods in order to choose all the key variables and remove features having low correlation with the label. The results indicate that the final model introduced in this paper achieved best among all other models as getting 96.77% score. The result in this paper benefits the related corporations in financial activities.