Optimizing Hyperparameters for Fraud Detection: A Comparative Analysis of Machine Learning Algorithms

This article aims to maximise the detection of economic transaction fraud by way of evaluating numerous machine learning techniques. The purpose is to optimise the model's performance with the aid of modifying its hyperparameters. The evaluation uses an anonymised dataset of credit score card transactions made by way of cardholders for the duration of Europe in 2023. It uses a methodical procedure that contains statistics preprocessing, visualisation, and evaluation. The model training strategies XGBoost, Decision Tree, Random Forest, Support Vector Machine, and Logistic Regression may additionally all have their hyperparameters changed with GridSearchCV. The efficacy of the models is mostly impacted by hyperparameter changes, even though the models’ algorithmic performance varies little. For an understanding of algorithmic power and computer efficiency, measures of the ROC curve, temporal complexity, and resource consumption are crucial. In the end, this study clarifies the intricate interaction that develops between machine learning algorithms and hyperparameter optimization in the context of fraud detection. Additionally, it provides practitioners with a useful guide for selecting and optimising models to maximise effectiveness in real-world fraudulent detection scenarios.