A Hyperparameters Tunned ML Algorithm for Fraud Identification in Banking and Financial Transactions

Abstract

Banking, a pioneering industry, is experiencing rapid growth, leading to the adoption of cashless transactions. Digital banking offers better service quality but has faced challenges from fraudulent activities. Since the banking industry is expanding quickly throughout the globe, using cash for payments is becoming less common. Instead, people are using cashless transactions. Digital banking customers receive higher-quality services in money transfers, cashless payments, credit cards, and prepaid cards. Nonetheless, the fraudulent activities of scammers have drawn attention to the security of digital banking, as a lack of adequate protection has discouraged many users from using the service up to this point. Even though fraud is not a new problem, its associated actions always cause billions of dollars' worth of annual losses to the world economy. Fraudulent actions carry a wide range of severe financial hazards that might jeopardize an economy's profitability and reputation. The study aims to introduce an efficient hyperparameter-tuned machine learning approach to detect fraud in banking and financial transaction systems. Proper preprocessing and application of feature engineering, such as outlier rejection, null value handling, standardization, and parameter tuning, have been incorporated with the approach. Later, the Extreme gradient boosting model was trained with tunned parameters and evaluated with test data. The model demonstrated praiseworthy performance, having 99.63% accuracy. Extensive analysis using feature selection, confusion matrix, roc, and tunning evaluation graph was conducted to detect fraud in financial transactions.