Online Payment Fraud Detection on Skewed Data: A Comparison Between KNN and Decision Tree

Online payments have become pervasive, making payment systems prime targets for fraudulent activities. This paper presents a contrasting evaluation of two ML algorithms, K Nearest Neighbors and Decision Tree, for predicting total losses due to fraudulent payments by processing a dataset of transaction details in.csv format. By analyzing patterns of fraudulent transactions, we utilize the dataset to identify key indicators of fraud and estimate the total losses provoked. The results illustrate the effectiveness of both algorithms in detecting fraudulent transactions and predicting the associated financial losses. We highlight each model's performance to lessen false positives whereas maintaining high detection accuracy through evaluation using standard fraud detection metrics. Thereby, it provides a robust framework for fraud detection based entirely on transaction data.