An Ensemble Learning Approach for Financial Fraud Detection: Combining Random Forest, K-Nearest Neighbors, and Gradient Boosting

Failure or frequent false alarms in fraud detection may trigger systemic financial risks, erode public trust, and even destabilize economic order, threatening national financial security. While traditional approaches relying on manual rules and basic statistical models demonstrate limitations in detection accuracy and efficiency, the emergence of sophisticated fraud techniques necessitates more advanced solutions. Using the Kaggle dataset, this study implemented comprehensive data preprocessing including feature selection, Synthetic Minority Oversampling Technique (SMOTE)-based class balancing, and categorical variable encoding. This paper compares seven machine learning algorithms, with Random Forest (RF) achieving superior performance (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0. 0 7 \%}$</tex> accuracy). Building upon this finding, this article developed a novel weight learning mechanism integrating RF, Gradient Boosted Decision Trees (GBDT), and K-Nearest Neighbors (KNN) predictions. The ensemble model demonstrated significant performance improvement, attaining 92.65% accuracy. Hyperparameter optimization revealed optimal configurations (5-20 estimators for RF, 90-100 neighbors for KNN), highlighting the importance of parameter tuning. The proposed framework capitalizes on the complementary strengths of constituent algorithms: RF excels in nonlinear pattern recognition, GBDT handles complex data structures, and KNN specializes in local feature analysis. This synergistic combination enhances prediction reliability and offers robust solutions for diverse fraud scenarios. The study's findings provide both theoretical insights and practical applications, establishing a foundation for nextgeneration fraud detection systems in financial risk management. In addition, the idea of the ensemble algorithm proposed in this paper can also be extended to other financial tasks.