Enhancing Online Transaction Security Through Comprehensive Fraud Detection Using Machine Learning

The purpose was to improve the effectiveness of Support Vector Machine (SVM) model using ensemble learning and hyperparameter tuning. Specifically, we utilized a stacking classifier that combines two base models: SVM and XGBOOST. SVM was optimized through key parameters which include the RBF kernel, and high regularization constant c value of 10, and Gamma ‘scale’ function for higher values of decision margins. For construction of the Model XG Boost was selected due to the reason that it is very accurate especially when it comes to imbalanced data sets and doesn’t require a high degree of customization as other algorithms but has a specific set several parameters like Maximum depth or learning rate which has to be adjusted. In the stacking ensemble, the meta-model was Logistic Regression which was then tuned to a greater number of iterations in order to achieve a better minimum. The meta-model was trained with cross-validation in order to avoid over fitting. The model was assessed with metrics that are commonly used, including accuracy, AUPRC and ROC-AUC. These metrics gave an indication on the ability of the model in classifying imbalanced data with high accuracy and predictive models. These techniques and hyperparameters adjustments brought a positive impact in the classification results confirming the success of the using a Hybrid Model composed by SVM and XGBoost. Here, a fusion of both technologies presents a direction for other machine learning applications that need accuracy and understandable results.