Financial fraud presents an escalating threat to the integrity of global economic systems, resulting in
billions of dollars in annual losses and undermining consumer trust in digital banking infrastructure.
Traditional rule-based detection engines fail to capture the subtle, evolving patterns of fraudulent
transactions and suffer from prohibitively high false-positive rates. This paper proposes a novel hybrid
data analytics model that integrates advanced machine learning architectures to systematically identify
fraudulent financial activities with high precision. The proposed framework combines an optimized
eXtreme Gradient Boosting algorithm for structured tabular data analysis with a Deep Autoencoder
network designed for unsupervised anomaly detection in highly imbalanced datasets. To resolve the
extreme class imbalance inherent in financial transaction data, an adaptive synthetic oversampling
technique is incorporated within the data preprocessing pipeline. The model was evaluated on two
benchmark public financial datasets containing over 284,000 credit card transactions and 6 million
mobile money transfers. Experimental results demonstrate that the hybrid model achieves an Area
Under the Precision-Recall Curve of 0.946 and a sensitivity of 92.3 percent, significantly outperforming
standalone conventional algorithms. The study concludes that blending supervised ensemble techniques
with unsupervised reconstruction error metrics provides a robust, scalable solution capable of securing
real-time financial networks against sophisticated fraud typologies.
