Data Mining and Fraud Detection on E-Commerce Transactions Using Comparative Analysis of K-Nearest Neighbors and Deep Neural Network

Abstract

This work focuses on creating and assessing machine learning models algorithms and methods for identifying fraudulent e-commerce transactions. Features including transaction type, amount, balance, and date are extracted from transaction data analysis and classified as authentic or fraudulent based on transaction limits and balance consistency. K nearest neighbours and deep neural networks train and test the dataset using standardized features. Once it has been divided into training and testing sets; the models' recall, accuracy, and precision are assessed. According to the results, the deep neural network model outperformed then K-nearest neighbours with an accuracy of 98.12% and recall of 83.52% comparing our result to other research works the deep neural network also outperformed results from other research which used fraud detection tools, such as Logistic Regression, random forest and SVM so on in terms. This is because deep neural networks can handle complex changes in data, and it also provides useful probabilistic insights, which makes it a powerful tool. The visualization of prediction probabilities is done using MATLAB 9.0 R2016a to improve interpretation and understanding. To further increase performance, future work will involve improving feature extraction, more real-world data sets and investigating more sophisticated models. This study shows how KNN and DNN may be used as machine learning models with an intuitive user interface to develop a reliable method for identifying fraudulent transactions and improving data mining and e-commerce transactions.