Credit Card Fraud Detection Using Lightgbm And Catboost

Abstract

The key objective of this research article is to evaluate the efficiency of LightGBM (LGBM) and
CatBoost algorithms when detecting instances of credit card fraud. Fraudulent credit card
transactions trigger losses of billions of dollars each year. Machine learning algorithms are used for
detecting credit card fraud. The credit card dataset which is available offline is used to assess the
effectiveness. Label encoding and one-hot encoding will handle each categorical feature in the
dataset. The data consists of attributes of different scales. Most methods of machine learning are
more efficient when the data attributes are of the same scale. MinMaxScaler is used to rescale the
data. On the scaled dataset, algorithms such as LightGBM (LGBM) and CatBoost are used to assess
accuracy levels. LightGBM improves the training speed and relatively maintains accuracy. CatBoost
reduces the time spent on parameter tuning because it provides great results with default parameters.
The results show that the deployed model achieved an accuracy rate of 97% and 98% respectively
using LGBM and CatBoost. The binary classification was implemented on a dataset to classify the
fraud and non-fraud cases.