A Novel Credit Card Fraud Detection by Outlier Identification and Elimination

Generally, credit card fraud refers to the unauthorized use of a credit card for accessing money; it
may result in financial losses. However, the existing studies didn’t detect credit card fraud based on
the transaction pattern similarity of various states. Therefore, this paper proposes SKCMA and
LHHS-ASV3N-enabled diverse transaction pattern similarity-aware credit card fraud detection.
Primarily, the “Credit Card Fraud Dataset” is taken. Then, the dataset is separated for training (80%)
and testing (20%). During training, the Apache Spark is employed to handle the big data. Then, the
outliers in the handled big data are identified and eliminated by utilizing EVA. Next, the attributes
are extracted from the outlier’s eliminated data. Subsequently, the unwanted attributes are reduced
by the technique named FG2DA. Thereafter, the data is grouped according to the state by employing
SKCMA. Based on the ADASYN, the grouped data is balanced. Lastly, credit card fraud is detected
as a fraud transaction and fraudless transaction by using LHHS-ASV3N. During testing, the outlier
in the 20% of the data are identified and eliminated. Then, unwanted attributes are reduced by
employing FG2DA. Based on LHHS-ASV3N, credit card fraud is detected. The experimental results
proved that the proposed technique achieved a high accuracy of 97.6%, thus outperforming the
prevailing methods.
Keywords: Adaptive Synthetic Sampling (ADASYN), Credit Card Fraud Detection, Outlier
Identification and Elimination, Attribute Reduction, Linear Horse Herd Scaling Based Artificial
Support Vector Nodes Neural Network (LHHS-ASV3N), Extreme Value Analysis (EVA), Big Data
Handling, and Apache Spark.