Detecting Money Laundering & Fraudulent Activities in Online Ordering Platforms

Abstract

" Since the advent of e-commerce, we have had access to online marketplace and restaurant platforms at the tip of our fingertips. Nowadays, online ordering platforms play a significant role in daily life, especially in meal provision amidst the COVID-19 pandemic. As a result, there has been a massive spike in demand for online ordering platforms. Despite all the advantages, online ordering platforms have created unparalleled opportunities for illicit activities such as money laundering and fraudulent activities. Detecting fraudulent financial transactions has become a priority for all financial institutes. Given the advances in modern technology and global communication, fraudulent activities have increased significantly, causing significant damages to countries‟ economies and society. Due to the limitations and weakness of the current rule-based fraud detection solutions, there‟s a high need for a more sophisticated machine learning-based framework to combat money laundering activities. This research paper aims to detect fraudulent transactions in a highly-imbalanced online ordering platform with an adequate model training time. A machine learning framework is fitted to an extensive dataset from the CAKE online ordering platform. The model is trained to foretell whether a new transaction should be listed as a fraudulent transaction or not, using background information about the card details, order amount and tip amount. The dataset is large and requires high computational power to process and train machine learning algorithms. Furthermore, another challenge for predicting whether a transaction is fraudulent or not is the highly-imbalanced distribution between the positive classes (0.1%) and the negative classes (99.9%). To handle the imbalanced class, the minority class data was oversampled using SMOTE and the majority class was under-sampled using random under-sampling. Computational effectiveness was achieved by the Apache Spark implementation, which provides distributed processing for big data workloads. Our experimental approach consists of several supervised learning algorithms for detecting fraudulent transactions. This research paper intends to produce, explain, and verify a machine learning model to determine which online ordering transactions should be flagged as possible money laundering and fraudulent transactions. "