{"id":1077,"date":"2026-08-16T21:47:28","date_gmt":"2026-08-16T21:47:28","guid":{"rendered":"https:\/\/els-journal.net\/wp\/about\/"},"modified":"2026-08-16T21:54:00","modified_gmt":"2026-08-16T21:54:00","slug":"30202-an-ensemble-deep-learning-approach-for-credit-card-fraud-detection-using-cnn-and-adaboost","status":"publish","type":"page","link":"https:\/\/els-journal.net\/wp\/?page_id=1077","title":{"rendered":"30202 An Ensemble Deep Learning Approach for Credit Card Fraud Detection using CNN and AdaBoost"},"content":{"rendered":"\n\n\n

Vol. 30, No. 2 – December 2026<\/h3>\n

An Ensemble Deep Learning Approach for Credit Card Fraud Detection using CNN and AdaBoost<\/h3>\n

https:\/\/doi.org\/10.53314\/ELS2630065S<\/h5>\n
Santosh Nirmal, Poonam Rakibe, and Pramod Patil<\/h5>\n
Abstract<\/b><\/h5>\n
Detecting credit card fraud remains a critical issue in financial security, demanding models that excel on imbalanced <\/span>tabular transaction data while balancing accuracy, interpretability,<\/span><\/h5>
and efficiency.<\/h5>
This paper proposes a novel hybrid CNN-AdaBoost framework, where convolutional neural networks extract intricate, <\/span>complex patterns from raw transactional features, and AdaBoost<\/span><\/h5>
performs robust iterative classification to refine predictions.<\/h5>
We evaluate the approach on standard credit card fraud benchmarks, achieving an overall accuracy of 95.43% alongside a <\/span>strong F1-score of 92.75%, surpassing baselines methods. Ablation <\/span>studies also conducted to confirm synergy: standalone CNN yields <\/span>93.12% accuracy, while pure AdaBoost reaches 93.58%, highlighting <\/span>how feature fusion addresses individual shortcomings in handling <\/span>noise and class imbalance.<\/span><\/h5>
The novelty stems from repurposing CNN\u2019s convolution layers to capture local, complex patterns directly from raw tabular transaction <\/span>data, feeding these rich representations into AdaBoost for <\/span>adaptive error correction. This hybrid uniquely blends deep feature <\/span>learning\u2019s nuance with ensemble boosting\u2019s efficiency and interpretability, <\/span>outperforming standalone methods on imbalanced <\/span>fraud detection. Prior works rarely fuse these for tabular domains, <\/span>yielding our 95.43% accuracy benchmark.<\/span><\/h5>
This provides valuable insights into the design of hybrid models for structured and tabular data classification tasks and establishes <\/span>a strong benchmark for future research in this domain.<\/span><\/h5>\n
Full text: <\/a><\/h5>\n\n\n\n\n\"cited<\/a>\n\n\n\n\nGoogle Scholar Citations <\/a>\n\n\n\n\n
<\/span> <\/center>