Detecting E-Commerce Fraud in Real Time with Machine learning

The rapid rise of online transactions spearheaded by e-commerce have restructured the global marketplace. Online transactions represent the highest level of convenience and efficiency, but have also posed a continuous threat of fraudulent activities that impact both customers and companies, and it is important to recognize that the rapid growth of e-commerce presents a problem of complexity predisposed to technological solutions to assist with amelioration. In this paper, we detail a systematic approach to engaging in fraud detection across many participants in an e-commerce transaction. The plan is based on building and validating several machine learning models to examine complex transactional data that considers multi-faceted dimensions. The classification machine learning models created include Random Forests, Support Vector Machines (SVM), Naive Bayes, Logistic Regression, and Gradient Boosting classifiers.Following the modeling, the classification schemes were unified in an experiment to examine each models’ accuracy in predicting between fraudulent and legitimate transactions. Our experience supports the precision regarding the random forest classifier (97.06% accurately classified). In this experiment we possessed a sizable dataset of transactional attributes, customer device and IP address details. The dataset was preprocessed utilizing techniques like standard scaling and one-hot encoding. Evaluation of models considered measures of Recall, accuracy, precision and F1 score on each models’ predictive abilities. Finally, we integrated the Random Forests classifier into a web application based on Flask