A hybrid Kmeans and ML Classification Approach for Credit Card Fraud Detection
Keshetti Sreekala, Rayavarapu Sridivya, Nynalasetti Kondala Kameswara Rao, Raman Kumar Mandal, G. Jose Moses, A. Lakshmanarao
Abstract
Credit card fraud poses a significant threat to financial systems and requires advanced methods for timely and accurate detection. This research explores a hybrid approach to credit card fraud detection by combining unsupervised learning, specifically K-means clustering, with five diverse classification algorithmsRandom Forest, Decision Tree, K-NN SVCand Naive Bayes. The methodology involves using K-means clustering to group transactions based on their features and subsequently treating the cluster assignments as pseudo-labels for training the classification models. The aim is to leverage the strengths of both clustering and classification to enhance the accuracy and efficiency of fraud detection.A dataset from Kaggle was used for experiments. Initially, elbow method applied to find the suitable clusters for the given dataset. Later, KMeans applied for performing clustering of dataset. Later, classification algorithms applied for evaluating performance of the model. The proposed method utilized various measures to measure the effectiveness of the hybrid approach.The results reveal the potential of combining clustering and classification for credit card fraud detection, with each algorithm contributing unique insights into different aspects of the data.