Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A robust machine learning framework for detecting temporal drift in financial fraud prevention.

Nikosi Zuberi1, Md Mostafizur Rahman2, Saeed Ur Rashid3

  • 1Department of Artificial Intelligence, European Institute for Materials AI & Technology, Kinshasa, Democratic Republic of the Congo. nikosi.zuberi@gmail.com.

Scientific Reports
|June 17, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Comparative Outcomes of Mini-Percutaneous Nephrolithotomy and Standard Percutaneous Nephrolithotomy for Renal Stones: A Prospective Comparative Study.

Cureus·2026
Same author

Direct Determination of the Interaction between Carbohydrate-Binding Modules and Lignin with Different Phenol Hydroxyl Contents.

Journal of agricultural and food chemistry·2026
Same author

Massive scabies outbreak in Rohingya refugee camps, Cox's Bazar: Epidemiology and impact of a mass drug administration (MDA) campaign - A retrospective study.

PLoS neglected tropical diseases·2026
Same author

Hybrid CNN BiLSTM architecture for smart grid cyberattack detection using smart meter data.

Scientific reports·2026
Same author

Tomosyn-2 Regulates Postnatal β-Cell Expansion and Insulin Secretion to Maintain Glucose Homeostasis.

Diabetes·2026
Same author

MaizeFormerX: a lightweight vision transformer with cross-scale attention for explainable maize leaf disease diagnosis.

Scientific reports·2026

This study introduces a robust framework for credit card fraud detection, optimizing machine learning models by carefully handling data imbalance and temporal shifts. An ensemble model achieved the highest performance, offering practical insights for real-world fraud detection systems.

Area of Science:

  • Machine Learning
  • Data Science
  • Financial Technology

Background:

  • Credit card fraud detection is challenging due to imbalanced data, non-stationary transaction patterns, and significant costs associated with missed fraud and false alarms.
  • Existing methods often struggle with data leakage and temporal drift, compromising real-world performance.

Purpose of the Study:

  • To present an end-to-end experimental framework for credit card fraud detection that strictly avoids data leakage.
  • To compare various oversampling strategies, classifiers, and ensemble methods under a rigorous chronological data split protocol.
  • To analyze temporal drift and its impact on model performance.

Main Methods:

  • A chronological split of the European transaction dataset into training, validation, and test sets was used.
Keywords:
Class imbalanceCredit card fraud detectionDeep neural networkEnsemble learningOversamplingPopulation Stability IndexSHAP feature selectionSMOTETemporal drift

Related Experiment Videos

  • Feature engineering (sinusoidal hour encoding, log-transformed amount) and SHapley Additive exPlanations (SHAP)-based feature selection were applied.
  • 12 classifiers, 3 MLPs, a deep neural network (FraudNet), and 7 ensembles were evaluated with 6 oversampling techniques.
  • Temporal drift was quantified using Population Stability Index (PSI), Kolmogorov-Smirnov (KS) tests, and Jensen-Shannon divergence.

    • Main Results:

    • The Soft Voting ensemble achieved the highest Matthews Correlation Coefficient (MCC) of 0.8060 and an AUC of 0.9703.
    • A specific Multilayer Perceptron (MLP) model without oversampling reached a high F2-score of 0.7722.
    • LightGBM with SMOTEENN demonstrated a substantial performance increase due to oversampling, while ADASYN consistently underperformed.

    Conclusions:

    • The proposed framework provides practical guidance for developing production-grade fraud detection systems that adhere to temporal and data integrity constraints.
    • Ensemble methods, particularly Soft Voting, show strong potential for credit card fraud detection.
    • Careful consideration of oversampling techniques and temporal drift is crucial for effective fraud detection model development.