Categorical Feature Encoding Techniques for Improved Classifier Performance when Dealing with Imbalanced Data of Fraudulent Transactions

Fraudulent transaction data tend to have several categorical features with high cardinality. It makes preprocessing complicated if categories in such do not an order or meaningful mapping numerical values. Even though many encoding techniques exist, their impact on highly imbalanced massive sets is thoroughly evaluated.
 Two datasets imbalance lower than 1\% of frauds been used our study. Six methods were employed, which belong either target-agnostic target-based groups. The experimental procedure has involved the use machine-learning techniques, as ensemble learning, along both linear and non-linear learning approaches.
 Our study emphasizes significance carefully selecting appropriate approach for machine algorithms. Using can enhance model performance significantly. Among various assessed, James-Stein Weight Evidence (WOE) encoders most effective, whereas CatBoost encoder may be optimal datasets. Moreover, it crucial bear mind curse dimensionality when employing like hashing One-Hot encoding.