Towards effective and robust bank fraud detection thanks to machine learning
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
With the increasing digitalization of our world and the digital storage of our information, we have witnessed a proliferation of fraudulent activities, of which credit card fraud remains the most prominent. Detecting credit card fraud is, therefore a major issue. Although extensive research has been conducted to counter credit card fraud, these frauds are increasingly evolving and complex to detect, hence the need for innovative solutions. Our solution aims to provide a robust and efficient fraud framework, capable of adapting to new threats such as adversarial attacks. We also studied the impact of data imbalance and adversarial attacks on model performance. To build our solution, we used two public datasets. The first step is to resolve the imbalance in our datasets using techniques such as Synthetic Minority Over-sampling (SMOTE) and Generative Adversarial Network (GAN). To further enhance the diversity of our datasets, we used two Autoencoders to generate more synthetic data. At this stage, we used five algorithms to test the performance of our models on both datasets, with XGBoost and Random Forest having the best performance. XGBoost has a recall score of 99.98% and an F1score of 99.98% on the first dataset, similar results are observed on the second dataset. The final step is to implement an adversarial attack, which resulted in a decrease on average of the recal performance metric of 15.77% on the first dataset and 20.22% on the second dataset. To counter this attack, we used the adversarial training method and found an average improvement of the recall score of 14.02% on the first dataset and 17.45% on the second dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it