Probabilistic XGBoost Threshold Classification with Autoencoder for Credit Card Fraud Detection

Due to the imbalanced data of outnumbered legitimate transactions than fraudulent transaction, detection fraud is a challenging task find an effective solution. In this study, autoencoder with probabilistic threshold shifting XGBoost (AE-XGB) for credit card designed. Initially, AE-XGB employs prevalent dimensionality reduction technique extract features from latent space representation. Then reconstructed lower dimensional utilize eXtreame Gradient Boost (XGBoost), ensemble boosting algorithm classify as or legitimate. addition AE-XGB, other existing algorithms such Adaptive Boosting (AdaBoost), Machine (GBM), Random Forest, Categorical (CatBoost), LightGBM and are compared optimal default threshold. To validate methodology, we used IEEE-CIS dataset our experiment. Class imbalance high characteristics reduce performance model hence preprocessed trained. evaluate model, evaluation indicators precision, recall, f1-score, g-mean Mathews Correlation Coefficient (MCC) accomplished. The findings revealed that proposed in handling able detect 90.4% recall 90.5% f1-score incoming new transactions.