Real-Time Deep Learning Networks for Fraud Detection in Modern E-Marketplace Systems
The project aims to develop new machine learning techniques and network architectures for detecting digital frauds, to address the increasing risks that businesses and their customers face as businesses digitize.
Target Sector: Finance
Lead PI: Assoc Prof He Bingsheng (NUS)
- Assistant Prof Bryan Hooi (NUS)
- Associate Prof Wong Weng Fai (NUS)
- Dr Carmen Cheh (Illinois at Singapore Pte Ltd, Advanced Digital Sciences Center, ADSC)
- Dr Chen Deming (University of Illinois at Urbana-Champaign)
- Chen Jia (Grab)
Host Institution: National University of Singapore (NUS)
Industry Partner: Grab
Digital transformation of businesses across various sectors has accelerated in recent years, especially during the covid-19 pandemic. As many more users participate in online activities, the risk exposure to frauds increases. According to tfageeks1, in 2019, Southeast Asia lost US$260 million to digital fraud. This number is increasing quickly over the years and can be far greater if unresolved.
In this project, we will develop a machine learning system to ease the development of new machine learning models for digital fraud detections in modern e-markets. Specifically, we will develop emerging learning techniques for detecting digital frauds: 1) emerging learning networks including LSTM (Long short-term memory) and GNN (Graph neural network) to capture the relationship of different entities and relationships in digital user behavior and transactions, and 2) one-class, few shot learning and reinforcement learning to address the unique challenges of data samples in digital frauds, and 3) online learning and hardware accelerations for timely adapting learning models to rapid changes in fraud behavior. We will partner with Grab, Southeast Asia’s leading super app, which offers essential everyday services to over 670 million people across Southeast Asia.
Potential Impact/Benefits to Target Sector
We will develop and deploy our model training and learning techniques into Grab services such as Grab food and Grab finance and demonstrate the effectiveness and efficiency of our techniques with Grab data sets as well as public data sets in fraud detection.
We will deliver publications and open-source systems, which the industry and ecosystem can benefit from. We intend to collaborate with more industry players and research how to adapt and advance state-of-the-art algorithms and systems to their applications and scenarios.