This paper is published in Volume-4, Issue-3, 2018
Area
Predictive Analytics
Author
Simranjeet Kaur
Co-authors
Sikander Singh Cheema
Org/Univ
Punjabi University, Patiala, Punjab, India
Pub. Date
14 May, 2018
Paper ID
V4I3-1399
Publisher
Keywords
Predictive Analysis, Imputation, Feature Scaling, Imbalanced Features.

Citationsacebook

IEEE
Simranjeet Kaur, Sikander Singh Cheema. Selective feature processing with k-Nearest Neighbor classification to predict credit card frauds, International Journal of Advance Research, Ideas and Innovations in Technology, www.IJARIIT.com.

APA
Simranjeet Kaur, Sikander Singh Cheema (2018). Selective feature processing with k-Nearest Neighbor classification to predict credit card frauds. International Journal of Advance Research, Ideas and Innovations in Technology, 4(3) www.IJARIIT.com.

MLA
Simranjeet Kaur, Sikander Singh Cheema. "Selective feature processing with k-Nearest Neighbor classification to predict credit card frauds." International Journal of Advance Research, Ideas and Innovations in Technology 4.3 (2018). www.IJARIIT.com.

Abstract

The predictive analytics are being used in many applications across the globe ranging from financial risk to avalanche studies. In this paper, a new approach is designed to predict the credit card frauds. This approach utilizes the imbalanced feature correction methodology, which eventually reduces the levitation of the features towards one class. The proposed model is designed to filter the credit card data by analyzing the multiple factors to analyze and predict the fraudulent transactions. The proposed model utilizes the maximum-minimum scaling method to scale the quantitative variables on 0-1 scale, after handling the missing values with column mean value. The SVM and KNN based classification method are used to predict the patterns for the credit card frauds. The experimental results have proved the proposed model based on SVM classification as the most efficient algorithm for the purpose of fraudulent pattern prediction. The SVM has been recorded with 99.94% (mean) of accuracy, which is slightly lower than KNN’s 99.95% (mean). Also KNN outperformed SVM on the basis of recall with (approx 91%) and F1 measure (approx 84%) against approx. 84.50% (recall) and approx 82.50% (F1 measure).
Paper PDF