Zindi is excited to introduce the winners of the #ZindiWeekendz Urban Air Pollution Challenge. In just 60 hours, the virtual hackathon attracted 254 data scientists from across the continent and around the world, with 126 placing on the leaderboard.
The objective of this challenge was to use weather data and daily observations collected from the Sentinel 5P satellite, which tracks various pollutants in the atmosphere, to predict PM2.5 particulate matter concentration (a common measure of air quality that normally requires ground-based sensors to measure) every day for each city. The data covered the last three months, spanning hundreds of cities across the globe.
In the beginning of the global pandemic, news articles stating that air quality has improved due to COVID-19 and the lockdown most countries went into started circulating, and although this is true for some locations, in parts of many African cities, air quality seemed to be getting worse as more people stayed home. For this challenge data scientists dug deeper into the data, finding ways to track air quality and how it was changing, even in places without ground-based sensors. This information will be especially useful in the face of the current crisis, since poor air quality makes a respiratory disease like COVID-19 more dangerous.
The winners of this challenge are: devnikhilmishra from India in 1st place, Team Covidata (karimcossentini, melkeor and plndz) from Tunisia in 2nd place, and Klai also from Tunisia in 3rd place. A special thank you to the 3rd place winner for their insights. This hackathon will be re-opened as a knowledge competition.
You can read more on how the Zindi community built a model that accurately predicts air quality in cities and towns across Africa on our Medium blog: Zindi solutions: A useful open-source model of urban air quality for Africa.
Helmi Klai (3rd place)
Zindi handle: Klai
Where are you from? Tunisia
Tell us a bit about yourself?
I am an engineering student in telecommunications at Enit and data science trainer at DataCoLab
Tell us about the approach you took.
I used an ensembling of two catboost models, one before adding aggregations features and the second after that. And this works very well with this data. I also used PCA for adding new features.
What were the things that made the difference for you that you think others can learn from?
I think I tried a lot of approaches and a lot of models before getting this accuracy. Therefore the most important thing that you should do with any AI problem, is to search more about the domain you work in and what the most important features are.
What are the biggest areas of opportunity you see in AI in Africa over the next few years?
Instadeep, Infor, cognira, expencia.
What are your thoughts on our winners' feedback? Engage via the Discussion page or leave a comment on social media.