How can we predict building claims costs in Small Medium Enterprises (SME) across Australia using advanced statistical machine learning techniques? Or group and price hundreds of different industries based on data and business knowledge? How should we deal with sparse data spread across hundreds of categories?
Commercial insurers are searching for solutions to these ongoing industry challenges, as well as young talents to contribute to their solutions. Dr Fei Huang describes the UNSW Business School’s Sandbox Project, which aims to do just that.
The Sandbox Project is a platform to connect industry, academics, and students, empowering them to co-create solutions to real-life problems. It introduces students to industry projects as part of their coursework. Industry partners and the lecturer co-develop and set project tasks that reflect contemporary challenges and then engage with student project teams throughout the term.
The initiative was launched by Dr Fei Huang as part of the Data Analytics Applications subject. Students were asked to complete a group assignment using a real industry problem developed in collaboration with Suncorp. It provided the 141 enrolled students with unique opportunities to learn, engage with industry, and contribute to the solutions to real-world actuarial analytic challenges. Students in this course have impressed us with a range of exciting and innovative solutions, which provides industry partners with fresh perspectives and creative ideas. The Sandbox collaboration experience shows that we can inspire talented young minds to play an important role in co-creating solutions to the pressing industry challenges through partnering with the industry.
The Sandbox Project with Suncorp
The Sandbox Project was intended to predict building claims costs across SMEs in Australia and suggest appropriate occupation groupings/clusters. The Australian Taxation Office defines a small business in Australia as an individual, partnership, company or trust that has an aggregated annual turnover of less than $10 million. This encompasses a diverse range of businesses that we interact with daily such as pubs, restaurants, hair salons and cinemas or the plumbers, cleaners and gardeners.
These businesses would usually require several types of insurance covers to protect them against the risks that can severely impact their operations. They can include building, contents, liability, business interruptions, theft, equipment, glass and even tax auditing!
This Sandbox Project focuses on building insurance cover which is one of the larger business lines for SMEs. Whilst not all businesses across Australia will require a building cover, many industries typically have a large exposure to risks related to their physical properties. It is important to understand the drivers of claim risk including electrical fire, water leaks, accidental damage, malicious damage and all of the natural disasters we may experience. The data used for this project are from small businesses across Australia between 2010 and 2015 provided by Suncorp.
The challenge – pricing difficulty of industries
The Australian and New Zealand Standard Industrial Classification (ANZSIC) is a hierarchical classification with four levels, namely Divisions (the broadest level), Subdivisions, Groups and Classes (the finest level). At the Divisional level, the main purpose is to provide a limited number of categories that provide a broad overall picture of the economy and are suitable for the publication of summary tables in official statistics. The Subdivision, Group and Class levels provide increasingly detailed dissections of these categories for the compilation of more specific and detailed statistics.
There are hundreds to thousands of different business types across industries such as accommodation, retail, food and manufacturing in Australia, (See ABS ANSZIC codes). This breadth is a key challenge faced by insurance companies as they look to price them. Often companies do not have enough experience (both knowledge and claims) across all industry codes to be able to group and price them appropriately. This means that there can be inconsistencies in pricing between similar industries causing adverse risk selection and unintended outcomes.
Bringing real-world industry challenges into university classes
Collaborating with the industry partner, Dr Fei Huang incorporated the industry challenge in the course as a group project assignment. Students were given six weeks to complete the Sandbox Project, working in groups of 4-5 students. This project was divided into three main tasks covering aspects of the course learning objectives of including exploratory data analysis, occupation clustering, and claims cost modelling.
The assignment structure was designed following the approach developed by Dr Kevin Liu. Students submitted intermediate milestone tasks and received feedback from peers, academics, and industry partners. This community of learning is designed to encourage students to be open-minded and achieve their best solution in the final submission. Students needed to demonstrate multiple skills through this project, including business understanding, data understanding, exploratory data analysis, modelling, evaluation and communication.
The Sandbox Project is a good example of problem-based learning (PBL), in which students learn about a subject by working in teams to solve real-life problems. In addition to being more engaging, it also develops skills in research and problem-solving, the ability to present ideas concisely and coherently, and to work in groups.
One pleasing aspect of the project is that students tackled the problem from different perspectives and with different tools, drawing on ideas they learned in the course and more broadly. For example, they:
- used modern R packages, such as tidyverse, to do data cleaning, visualisation and processing;
- applied advanced statistical machine learning techniques, such as GLM with shrinkages, GAM, gradient boosting machine, random forests, and neural networks to model the claim costs and compare the outcomes; and
- used RMarkdown to generate reproducible technical reports.
Video submissions were the main outputs of the Sandbox Project and students showed their strong ability to present ideas clearly and precisely to the audience.
An undergraduate student group consisting of Yingkun (Queenie) Huang, Michael Jacinto, Kevin Li, Harry Peng and Melantha Wang were selected as the top group. A video link of their project presentation can be accessed here. The team applied k-means clustering to classify risks by grouping occupations with similar risk levels. Additionally, their team compared various machine learning algorithms to predict claims costs, including GLM, XGBoosted GLM, GAM and XGBoost on fire and non-fire data separately.
“Inspired by Section 7 of Ferrario, Noll, and Wüthrich (2020), we proposed an innovative combined model, called XGBoosted GLM, which used a GLM as the backbone, and overlaid an XGBoost layer to explore any structural patterns in the residuals. This model took advantage of the interpretability of GLM and prediction accuracy of XGBoost, and overcame some of the technical limitations of GLM (Yao, 2013),” the team explained.
All models they tested identified building sum insured and region risks as the most significant and important predictors of claims cost. And their proposed innovative model performed the best across all the candidates they tested.
The Sandbox Project created an authentic learning experience for students that enhances their learning and employability skills.
Queenie Huang explained that “the Industry Sandbox Project provides me with great opportunities to apply the academic knowledge to real-world applications. By collaborating with friends from different disciplines, I’ve learnt to balance the business objectives and constraints, and come up with a tailored solution.”
Melantha Wang highly valued the opportunity to apply techniques learnt in class to a real business challenge.
“Unlike most other uni assignments where we work in a theoretical frictionless world, here we really had to get our hands dirty in the data and think through all the practical aspects of the problem (e.g. feasibility of the solution, what are the risks involved),” Melantha explained.
Michael Jacinto believed the Sandbox Project is a ground-breaking innovation in educational design.
“What stood out to me about the Sandbox model was how it truly offered a learning experience. One thing I often hear from university students is that they crave genuine feedback and an opportunity to apply that feedback and learn from their mistakes – something many university courses do not enable.
“Overall, the Sandbox model was a fun and engaging learning experience like no other that brilliantly capped off my undergraduate studies,” Michael said, “I think the Sandbox should be a model for future assignments in not just Actuarial courses, but across all university areas.”
Industry partner’s experience
The Sandbox Project allows leading industry partners to contribute to the development of the next generation of actuarial students through direct input into curriculum design and active involvement in challenge-driven education. The program also creates greater visibility and empowers industry partners gain direct access to bright and talented students while earning fresh perspectives and creative solutions to current problems
Jason Yu, Pricing Manager, Suncorp, said “I believe the Sandbox project was an innovative initiative that took the students out of their comfort zone.”
“Firstly, having access to industry-like datasets (no more data with perfect distributions) and having a real-life business problem to solve challenged the students to contextualise and self-research an industry they are completely unfamiliar with,” Jason commented.
“Secondly, the milestone approach to the assignment throughout the course of the program mimics a pitch, plan and execute type of proposal that is more closely aligned with how projects in business operates.”
Lily (Siying) Liu, Pricing Analyst, Suncorp, said “I really enjoyed working with you and students on this project. I find it is well-organised and well-delivered, especially how the assignment was split into different segments and how industry partners were involved to provide information and feedback.”
“I personally enjoyed the engagement and involvement as an industry partner, especially by giving an introductory information session, middle check-in session and final sharing session really helped to interact and engage with the students,” Lily commented.
The Sandbox Project provides opportunities to academics to develop meaningful relationships with the industry for continuous innovation in education and research.
Dr Fei Huang believes this is a very successful innovation and attempt in actuarial education. Actuarial studies is an applied discipline, and traditionally there is a gap between university education and real-world practice. Through the Sandbox Project, we bridged the gap via the collaboration of three parties, students, academics, and industry partner. It is great to see that students were very engaged to work in a team and dedicated to completing each part of the project to a great extent.
Impact of the Sandbox Project
The successful implementation of the Sandbox Project with Suncorp has also been recognised internally and externally.
The highest-scoring student group was invited to present to Suncorp teams of their approach to tackle this industry challenge. Melantha, now an Honours student at UNSW, was inspired by this Sandbox idea and has decided to do her Honours project closely linked to this industry challenge using advanced machine learning techniques.
“Over the course of Data Analytics Applications, the students were able to apply techniques learnt from the course, utilising both industry and publicly available data to derive insights aligned towards business objectives,” Jason commented. “A big congratulations to the top student group for the amazing work they’ve done.”
“It is great to see the Sandbox Program continue to be rolled out across different courses in the Business School,” Dr Kevin Liu, Undergraduate Actuarial Program Coordinator and Deputy Director of the UNSW Sandbox Program said. “It is fantastic to bring together academics, students and the industry to empower career-focused learning at scale, co-create solutions for contemporary real-world problems, and build long-lasting university-industry partnerships through education and research collaboration.”
Professor Bernard Wong, Head of School, Risk and Actuarial Studies, UNSW Sydney, commented “The UNSW School of Risk and Actuarial Studies’ strategic goals revolve around the pillars of education, research, and societal/industry impact and engagement. The Sandbox Program stands at the nexus of the three pillars, and allows students, industry partners, and academics to come together as a community and collaborate with each other at scale. This model is an approach that we are implementing across a number of courses in the UNSW Actuarial Studies program, and, indeed within the UNSW Business School more broadly.”
CPD: Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.