Registering a child for his or her first year of school weighs heavy on the minds of every parent in Singapore. The process is competitive, with many having to ballot for entry at various stages of the registration process to the most popular schools in Singapore.
The key challenge to a parent is: Should I register my child at School A, forgoing my chance to register in School B – considering that School A might turn out to be oversubscribed and balloting might not end in my favour?
As the chance of balloting increases with subsequent phases, missing out the chance to register for School B earlier on might mean losing the chance entirely later on.
We provide a solution to a big pain point for parents by giving them an idea early in the registration process of the chances of their child getting into the school they want.
What makes this project innovative?
Machine learning: While past exploration of this topic often focuses on the anecdotal discussion of which schools are more likely to ballot, we cut straight to the point by answering the million-dollar question: What is the over-subscription risk for any given school in a particular phase of registration? Comparison widget: Our simulation tool includes a comparison mode so parents can - at a glance - see the relative over-subscription risk for two different schools. This helps them consider the trade-off between taking their chance on one school over another. Expert analysis: Information is also presented Q&A style and senior education correspondent Sandra Davie also writes on how to choose a primary school for your child.
What was the impact of your project? How did you measure it?
This popular graphic goes beyond providing information to readers in a typical article format to become a valuable, reusable widget for parents to get an idea of the chances of their child gaining entry to a school of their choice at each phase of the registration process. Average time on page during the ﬁrst 2 months of the registration phases is almost 5 minutes, more than double the website's average of about 2 minutes. This project won a Bronze in the Society for News Design’s Best of Digital Design 2019 competition.
Source and methodology
We collect data from the past six years of Primary 1 registration exercises, and build a machine learning model that predicts the number of applicants by school and by phase, given the number of vacancies in that phase is known. Outcomes of this prediction (number of applicants vs vacancies) are used to determine if balloting will be required in that phase. We repeat the prediction using bootstrapping to simulate many different outcomes. For example, if in 100 simulations, 33 ended up with balloting (applicants > vacancies), the likelihood of the school going into balloting in that phase will be indicated as 33%.
We used Python scikit-learn library to model our data and perform machine learning. Simulation is implemented as a microservice that is consumed by our front-end widget. This makes the widget a plug-n-play tool that can be shared across applications built by our sister publications.
- THONG Yong Jun - LEE Pei Jie - Ravi GANDHI - Hannah YAN - Jocelyn TAN - Rodolfo PAZOS - Amelia TENG