Abstract:
"The rush in digital transactions has underscored the importance to develop accurate and
trustworthy mechanism for detecting online payment fraud. In response to these challenges,
author propose an innovative hybrid machine learning approach. The methodology integrates
different algorithms to enhance the accuracy and resilience of fraud detection systems. The
process initiates with the curation and preprocessing of a comprehensive dataset containing
both genuine and fraudulent transaction data. Subsequent feature engineering extracts crucial
insights from raw data, augmenting the model's discerning capacity. Random Forest classifier,
Gradient Boosting classifier, AdaBoost classifier, and K-Nearest Neighbors classifier (KNN
classifier) were judiciously chosen to develop an ensemble of model as an output of the
research.
The main point of the methodology lies in the ensemble model creation phase, where
predictions from individual models merge through voting mechanism leveraging the unique
strengths of each classifier algorithm. This merging results in a more accurate and adaptable
fraud detection system. Critical cross-validation and hyperparameter tuning are used to
meticulously refine the ensemble model, ensuring optimal performance without overfitting.
The ensemble model demonstrated superior performance, significantly improving fraud
detection rates while reducing false positives, thereby enhancing user experience. In summary,
the proposed hybrid machine learning approach offers a strong and precise solution for online
payment fraud detection. By using the strengths of various classifier algorithms, it strengthened
accuracy and establishing a robust defence against the evolving landscape of digital payment
fraud strategies.
Subject Descriptors:
• Computing methodologies → Artificial intelligence → Machine learning → Hybrid
learning approaches
• Applied computing → E-commerce → Security and privacy in e-commerce → Fraud
detection
Keywords: Machine learning, K-Nearest Neighbors classifier, Random Forest, Logistic
Regression, Ensemble model, Voting, Cross-validation, Hyperparameter tuning"
