Finding That One in a Million

Prescribed by Math

A term first coined by IBM in 2010, prescriptive analytics is a hot topic today. Considered the final stage in the analytics evolutionary path, it aims to optimise decision-making by analysing past data and predicting unknowns to determine the best course of action forward. A research programme spearheaded by the Institute of Operations Research and Analytics at NUS seeks to advance this process further.

"Intelligence consists not only in knowledge, but also in the ability to apply knowledge in practice.” The Greek philosopher Aristotle might have lived 2,000 years before the age of big data, but his words ring especially true today. Data-mining technology has placed a wealth of information at the fingertips of many – from corporations and individuals to policymakers and business leaders. But what does one do with all that knowledge? 

Enter prescriptive analytics. To trace its roots, one might have to go back to 1950, when the ANIAC computer generated the first weather forecast models. By the 1980s, Decision Support Systems that gather and analyse data were already being applied to operations, financial management and strategic decision-making at multiple levels. Fast forward to 2020, and analytics has evolved beyond the descriptive – seen as the first step, which focuses on using historical data to provide a context for understanding information and numbers – and predictive, which uses current and past data to project possible outcomes of the unknown. “Prescriptive analytics goes a step beyond,” says Professor Teo Chung Piaw (Science ’90), Director of IORA, and the lead principal investigator (PI) for the new project. “While predictive analytics tells you if it might rain tomorrow, prescriptive analytics tells you to bring an umbrella.”

Taking on complex problems

While prescriptive analytics has been around for some years, it is a field of constant change: “It is an extension of the current movements to make better use of the data – it is an evolution,” explains Prof Teo. And this new research project is advancing that evolution. 

Professor Toh Kim Chuan (Science ’89), Professor in Science at the Department of Mathematics and recipient of the 2019 President’s Science Award — also a lead PI of the project in charge of creating the solver technology to tackle complex problems — elaborates further. He explains that the research takes prescriptive analytics beyond the current tools that largely leverage on linear programming models. Such models are better suited to prescriptive problems where the variables are directly correlated – such as, the relationship between number of man-hours required and the total production cost of a commodity. It also goes beyond model-free approaches that exploit the advancements in machine learning to crunch big data and better understand the trade-offs. Instead, it leverages on IORA’s expertise in solving large-scale conic programming, mathematical optimisation and operations research problems to develop state-of-the-art solutions for complex problems. 

The project takes a three-pronged approach: developing new algorithms/solver technology; creating platforms for the implementation of the technology; and finally, applying it to real-life problems, which Prof Toh describes as challenges “with millions of scenarios to consider, and where the relationship between different variables is not linear.” “Real-life problems have many unknowns – and decision-making goes beyond a probability theory,” says Prof Teo. He illustrates this with the example of an army going to war: “The commander decides on the course of action to take based on the intelligence gathered on the enemy: what they are doing now, and how they might react. This is scenario planning. However, scenario planning is often just a thought exercise with scenarios constructed in a very ad hoc manner. Our challenge is in replicating that thought process, and making it better by anticipating the millions of scenarios and how they will each pan out, based on the internal and external reactions,” explains Prof Teo, who was recently named a fellow of The Institute for Operations Research and Management Sciences (INFORMS) for his stellar contributions in optimising business processes. He is the only fellow named from an Asian institution in the organisation’s history. “The objective of the research is to come up with a systematic model that can be applied by decision-makers in different fields,” he says.

The A-Team

This ambitious five-year research programme is a multi-disciplinary effort. Apart from Prof Teo and Prof Toh, other PIs include Professor Melvyn Sim (Engineering ’95), Professor and Provost’s Chair at the Department of Analytics & Operations, NUS Business School; and Professor Andrew Lim Leong Chye, who heads the Department of Industrial Systems Engineering and Management, Faculty of Engineering. Highly respected in their own fields, the four form the core of a 10-person operation for this multi-disciplinary programme. 

Yet, this is a project that takes more than 10 persons to realise. “Most of our research is tested with partner companies where our data comes from,” says Prof Teo, who reveals that these range from regional corporations to Singapore-based enterprises. “While ours is an academic research, our tools still need to be tested. So we use what we develop to solve real-world problems – such as helping a car-sharing service optimise their vehicle deployment, and at the same time demonstrate how our model adds value.” He adds that in this respect, conducting this research programme in Singapore has its advantages: “With the Smart Nation initiative, companies are more receptive to sharing data and there is a willingness to participate. Our country already has the infrastructure for such collaborations.” And while the current funding is for a five-year programme, their vision for it goes way beyond. “Right now, only researchers in the community are applying the science of prescriptive analytics. The challenge is to push it into the classrooms, the business schools and engineering schools, and we hope to do this by demonstrating its usefulness,” says Prof Teo. “If our body of work can be integrated into classroom teaching, our students can then go out to deploy some of these models in the public sector.”

One thing is certain: this goal will take diligence and patience to achieve, thus the team is actively pursuing partners who can help realise this long-term vision. “For the programme to have tangible, visible impact, it will take at least 10 years. But we see this as integral to developing Singapore as a smart nation,” says Prof Teo. “A nation isn’t smart because it has smart technology, but because it has smart people. And we are trying to make people smarter.”