KUALA LUMPUR, 28 May 2020 – GBG (AIM:GBG), the global technology specialist in fraud and compliance management, identity verification and location data intelligence, today announced its expansion of AI and machine learning capabilities for its transaction and payment monitoring solution, Predator, making deep learning and predictive analytics available to their entire digital risk management customer journey. GBG first announced its machine learning capabilities for Instinct Hub, their digital onboarding fraud management system in January this year. The new AI capability additionally processes third party data – device fingerprinting, geolocation, mobile and IP, endpoint threat intelligence, behavioral analytics – assimilated into the GBG Digital Risk Management and Intelligence platform to enhance their model performance in fraud detection.
With the current pandemic giving rise to changes in consumer behavior in spending, fund transfers and loans, the ability to re-learn new data and adapt to new environments can help financial organizations detect emerging and escalating transaction and payment fraud trends and mitigate fraud loss. Based on GBG’s “Understanding COVID-19 Fraud Risks” poll results in April, 37% of respondents see transaction fraud as the fraud typology that they are most vulnerable to.
“Fraud is irregular, complex and evolves dynamically. Standard fraud model deteriorates over time, exposing businesses to new fraud typologies and fraud losses. Through continual and autonomous model training in GBG Machine Learning, we address the issue of model deterioration,” said June Lee, Managing Director, APAC, GBG.
“Today machine learning provides an average of 20% uplift in fraud detection, GBG Machine Learning has performed well to provide incremental alerts on missed frauds for our customers,” adds Lee.
GBG Machine Learning utilizes Random Forest, Gradient Boosting Machine and Neural Networks – three leading and proven algorithms for fraud detection. These algorithms embody strong predictive analytics, fast training models and high scalability, learning through both historical and new data. GBG AutoML (Automated Machine Learning) enables adaptive learning to provide the model capability to re-learn and update itself automatically based on a specified time interval.
“Through our APAC COVID-19 fraud risk poll results, digital retail banking services are growing in demand, from e-wallet, e-loan, digital onboarding, to digital credit card application; most respondents see a rise in e-banking services utilization. The ability to easily spot complex fraud and misused identities in first party bust outs and mule payments, high volume and high velocity frauds such as online banking account takeover and card not present frauds across both onboarding and ongoing customer payments becomes more pressing today,” said Michelle Weatherhead, Operations Director, APAC, GBG.
“In addition, segments like SME lending and microfinancing would be able to harness machine learning to spot irregularity in borrower patterns by assimilating both identity, profile and behavioural type data. GBG Machine Learning is able to analyse large sums of data using algorithmic calculations on multiple features to determine fraud probability in greater accuracy,” quips Dr Alex Low, Data Scientist, GBG.
GBG Machine Learning is designed to simplify machine learning deployment for both fraud managers and data scientists, removing the need to have a data scientist in-house or having to work back to back with the vendor to lower cost of operation. The solution offers high user controls from feature creations, model selection and configuration, results and analysis interpretation and alert thresholds. Users can also configure the solution to auto schedule and update new fraud patterns through its intuitive user interface with tool tips built in.
The solution takes a “white box” approach to provide an open and transparent modelling process for ease in model governance and meeting regulatory requirements. The machine learning score and top contributing features to results are visible to the users who need to gather further insights and understanding on the machine learning model performance and behaviours.
