In November 2019, I completed my first Hackathon at the Microsoft Reactor in London. The Hackathon was organised by UCL’s Data Science Society in collaboration with Microsoft and American Express and the theme of the event was about leveraging data science for generating insights in the credit card industry. Particularly, within the competitive credit card industry, the main objective is for companies to expand and retain their customer base. However, issues like high delinquency and default rates, growing subprime lending volumes and growing demand for product personalisation are obstacles to this end-goal. We were assigned into teams and our mission was to use data analytics tools to study various datasets from AmEx, analyse where they were performing poorly and pitch a business solution. As it was a Hackathon, we had to come up with this pitch over a brief 8-hour period.
Another valuable skill I learnt was using Microsoft Azure, a cloud computing platform created by Microsoft. Using Azure hosted notebooks, our team created various data visualizations and implemented machine learning models to understand customer credit card usage and spending patterns. We were able to deliver a coherent pitch describing our learnings from the data, however given the brief time, I feel we weren’t able to arrive at any specific conclusions or generate relevant business insights. Nevertheless, watching other groups present their pitches and seeing their technical solutions was quite inspiring as it allowed me to appreciate other perspectives and learn about other ways to use machine learning tools.
Here are my 4 biggest takeaways from this competition:
- Domain Knowledge is the important part of data science and ML research! The insights we generated was just as good as our understanding of the business problem. That being said, had we prepared more on understanding the problem, I feel we could have performed far superior analyses.
- Storytelling is what data science essentially is. Recalling our work process during the hackathon, I feel we made the mistake of immediately programming and creating the algorithms, rather than planning what questions we wanted to ask/research and how we wanted to present the story. If given the opportunity to change this workflow, I would have spent more time on the actual planning process.
- Simplest algorithms are the best sometimes! Normally in machine learning, it is thought that the more complex the model, the better it will be. Just looking at the best teams during this Hackathon, I noticed they all just used simple algorithms like decision trees and logistic regression to predict customer attributes. This taught me that starting simple is always the best way to approach a task.
- Presentation is key. At the end of the competition, we had to present a pitch to a panel of experts who were not specialized in statistics/data science. In the data science field, very often we will be discussing ideas with people who are specialized in other domains. Hence, presenting the ideas clearly is an important skill which I learnt through this competition.