This project was produced during the 2019 Canadian electoral campaign. It allows you to build a character (gender, age, province, income, religion, etc.) and to visualize which political party is more likely to get the vote of the character you created. To calculate this probability, we fed a machine learning algorithm with the answers of 387,671 participants of an online survey.
The illustrations are made in SVG. For the machine learning algorithm, we partnered with the non-profit Vox Pop Labs, which also in charge of the online survey gathering the data. The algorithm itself was based on three models: one multinomial logistic regression, one support vector machine, and one extreme gradient boosting.
What was the hardest part of this project?
Our team has to produce projects that work perfectly on both desktop and mobile. Since the user has to build its voter, there’s a lot of interactions involved. The hardest part was to find a design that would let the user make the choices easily, without friction, on the small screen of a phone. We based our design on the height and width of an iPhone 5, which is an old device, with a very small screen. We thought that if it worked on an iPhone 5, it would work on any screen.
What can others learn from this project?
With Artificial Intelligence (or machine learning), an amazing new set of tools becomes available. But it is sometimes hard to find a journalistic purpose for new technologies. We hope this project will give ideas to other teams to apply machine learning to concrete journalistic projects.