Royal Roads Super Sleuths Use AI to Ferret Out Financial Crimes
Dr. Mark Lokanan and his research team at Royal Roads University are pioneering the use of advanced artificial intelligence (AI) models to expose financial fraud. By harnessing AI’s capabilities, they sift through vast troves of data to spot abnormal transactions that warrant further investigation by forensic investigators in banks and financial institutions.
The field of AI is already well-established in commercial fraud detection. However, what sets Lokanan’s work apart is the development of customized AI models to target specific fraud types, ensuring precision in flagging deviations amidst extensive datasets.
AI’s prowess lies in its ability to emulate human intelligence, undertaking complex tasks such as reasoning, decision-making, and problem-solving. In fraud detection, Lokanan’s models are finely tuned to identify deception or misuse patterns that deviate from the norm.
A financial statement, for instance, typically adheres to a standard flow. The innovative models crafted by Lokanan and his team are adept at pinpointing anomalies within these financial documents. They are trained to recognize unusual patterns, such as unexpected spikes in company spending or dubious data manipulation.
Operating in real-time, the AI models alert users by notifying them on their computer screens when a deviation is detected. This prompt enables trained in-house investigators to delve into the irregularities to determine if fraudulent activity is involved.
Lokanan’s approach involves using machine learning and deep learning—sophisticated AI techniques that rely on algorithms honed through meticulous testing. “There’s a lot of math behind this,” Lokanan explains, highlighting the complexity of the underlying equations.
The AI models developed are designed to identify specific anomalies, shedding light on particular fraud types based on an analytical framework that directs how fraud should be scrutinized. While the algorithms supporting these models aren’t novel, the models’ application is indeed groundbreaking.
To date, Lokanan’s models have undergone rigorous testing on two extensive datasets—one focusing on financial statement fraud and the other on supply-chain fraud. The financial fraud dataset, procured from the University of Southern California, includes data on about 1,000 companies examined by U.S. officials. Conversely, supply-chain fraud data was freely accessible to the public.
The financial fraud models target credit-card fraud and mortgage fraud, instances where false information is furnished to acquire properties. For supply-chain fraud, the team delved into just how individuals might exploit the ordering process for personal gain.
For example, fraudulent activities could involve ordering a product and subsequently claiming non-receipt, or canceling orders once shipments are underway. AI models are adept at identifying these suspicious activities, flagging instances of delayed deliveries or canceled shipments for further company review.
The models are capable of generating periodic reports, such as weekly updates, tailored according to a company’s specific requirements. However, these models are still under development and not quite ready for market release, as ongoing research is focused on their robustness and accuracy.
So far, Lokanan is pleased with the models’ performance in identifying red flags in financial and supply-chain fraud data. The models have undergone peer-review, a process spanning up to two years, with results published in journals. Notably, supply-chain fraud models achieved accuracy scores in the mid-90s out of 100, while financial data models scored in the mid-80s.
Interest in Lokanan’s work is growing, with requests for collaboration from professors globally. Such advancements are timely, given reports like the Canadian Anti-Fraud Centre’s 2022 annual update, which highlights escalating fraud-induced losses—rising to $530.4 million from $383 million in 2021.
Beyond financial crime, Lokanan envisions broader applications for his research. The methodologies employed to detect fraud could be adapted across various sectors, with plans already underway to develop models predicting employee retention—a crucial metric given the high costs associated with turnover.
His ambitions extend to addressing temporary worker fraud—a significant issue within Canada. Using data from Statistics Canada, Lokanan aims to scrutinize compliance among companies, particularly concerning accurate wage reporting and worker classification.
Moreover, Lokanan is keen to analyze penalties imposed by regulatory bodies, assessing whether fines proportionalize with offenses.
For inquiries, contact: [email protected]
To comment on this article, please write to the editor: [email protected]
