Transparent Peer Review By Scholar9
CREDIT CARD FRAUD DETECTION USING SUPPORT VECTOR MACHINES AND NAÏVE BAYES (SVM-NB)
Abstract
in the financial business, financial fraud is an ever-increasing threat with far-reaching implications. Data mining is crucial in detecting credit card fraud in live transactions. The work investigates the performance of Naïve Bayes, Support Vector Machines and Hybridization of these two techniques on a highly imbalanced dataset (credit card dataset). Credit card fraud dataset is sourced from the European Cardholders that has 284,807 transaction records. The work was implemented in python. The performance of the model is evaluated using popular metrics namely; Accuracy, Sensitivity, F1-Score, Prevalence, Precision, Specificity, False Alarm Alert and Balanced Accuracy. The results showed based on average accuracies were 99.80% for Support Vector Machine, 98.0% for Naïve Bayes and the Hybrid model produced 99.99%. The comparative result showed that, there was an improvement in the Hybrid model.
Archit Joshi Reviewer
02 Oct 2024 05:51 PM
Approved
Relevance and Originality
The article tackles a critical issue in the financial sector: the increasing threat of credit card fraud, making it highly relevant. By focusing on data mining techniques such as Naïve Bayes and Support Vector Machines, the research contributes original insights into effective fraud detection methodologies. The exploration of hybrid models adds further originality, addressing gaps in existing literature on fraud detection.
Methodology
The methodology section outlines the use of a substantial dataset sourced from European cardholders, which includes 284,807 transaction records. However, more detail on the dataset's composition, such as the ratio of fraudulent to non-fraudulent transactions, would enhance understanding. While the implementation in Python is mentioned, further specifics regarding the data preprocessing steps and model training procedures would strengthen the methodology's robustness.
Validity & Reliability
The article reports impressive accuracy rates—99.80% for Support Vector Machines, 98.0% for Naïve Bayes, and 99.99% for the Hybrid model—suggesting strong model performance. However, the validity of these results could be bolstered by providing details on cross-validation methods or any testing against unseen data. Addressing potential biases within the dataset and evaluating the models on different datasets would enhance reliability.
Clarity and Structure
The article is generally well-structured, guiding the reader through the problem statement, methodology, and results. However, clearer definitions of key terms and metrics used in evaluating model performance would improve comprehension. Incorporating visual aids, such as graphs or tables comparing model performances, could enhance clarity and provide a more engaging presentation of results.
Result Analysis
The results indicate that the hybrid model outperforms the other techniques, which is a valuable finding. However, the analysis should delve deeper into the implications of these results for real-world fraud detection applications. Discussing the limitations encountered during the study, such as the effects of dataset imbalance and overfitting, would provide a more comprehensive understanding. Recommendations for future research, particularly exploring additional techniques or the application of models in different contexts, would encourage further investigation in this area.
IJ Publication Publisher
Ok Sir
Archit Joshi Reviewer