Meet the Winners of the Layer.ai Air Quality Prediction Challenge
Meet the winners · 22 Dec 2022, 12:35 · 5 mins read ·
9

Meet Team 2 And a Half Id!ots! (Nkosana_Daniel and Robert_Selemani), winners of the Layer.ai Air Quality Prediction Challenge. The challenge attracted 662 participants, all vying for a $3000 prize pool. The objective of this challenge was to predict air quality readings from AirQo’s sensors using Sentinel 5P data.

Nkosana Daniel Mlandu, Zimbabwe

Please introduce yourself.

My name is Nkosana Daniel Mlandu (Nkosana_Daniel). I'm currently a 4th-year Engineering Student at the National University of Science and Technology, Zimbabwe.

Tell us a bit about your solution and the approach you took.

We spent a decent amount of time trying to understand the data as well as possible, from the data-generating process to the missing values and our ‘hunches’ about the quality of the data. We also performed adversarial testing, where we built a simple random forest classifier to tell apart the train and test data. We quickly observed that the test data was out-of-distribution and that the degree of difference was rather too severe. We also observed that the choice of imputation technique affected the degree of difference in distributions, hence we were able to select the optimal technique from the ones we had tested.

Please find a detailed write-up of our solution posted in the discussion forum here.

What set your winning solution apart from others?

In this competition we learnt quite valuable lessons. Understand the data, ask questions, make hypotheses and test them out by running experiments. And most importantly, it's cool to spend a lot of hours coding and debugging models, but it's fine to take a short break during a competition, it'll certainly give you the clarity of mind needed to generate new ideas.

How do you prepare for a challenge?

I start by reading the problem statement and try to understand what problem the challenge seeks to solve. I always make sure to go through the discussion forums in case someone has posted something about the challenge, whether it's about missing data or starter notebooks. Most times it'll give you an idea of what others have tried, what's working and what's not working. From there I try to build my solution. The pipeline is a bit different from competition to competition but it's mostly Exploratory Data Analysis (EDA) > Data Cleaning > Feature Engineering > Modeling > Evaluation and a lot more of going back and forth, iterating.

Words of encouragement for others, or advice that has helped you?

Having a good validation strategy will go a long way in testing your experiments, and also help to avoid leaderboard shake-ups.

I wouldn't advise on looking for the best model for a particular competition or having a ‘till death do us part’ kind of attachment with a certain model architecture. It's surely a good starting point to search for the most appropriate model, but don't let it limit you; DON'T BE RIGID. It's data science so you need to experiment with different things and see what actually works. Fall in love with the science, and enjoy the process.

Don't forget to track your experiments! Have a systematic way to track all the things you've tried and what has worked so far. BE SYSTEMATIC AND INTENTIONAL!

What do you like about Zindi?

The ML ecosystem in Africa is quite different from other parts of the world, we don't have many open source datasets. For the majority of ML enthusiasts, the closest you can get to working with real African datasets is through Zindi. It's also a good platform to build your portfolio. Zindi has allowed me to network with like-minded individuals and learn from them as well.

Robert Selemani, Zimbabwe

Please introduce yourself.

My name is Robert Selemani (Robert_Selemani). I'm currently a Data Scientist at Repliteq AI, and Zindi Country Ambassador for Zimbabwe. I am based in Harare, Zimbabwe.

What set your winning solution apart from others?

Understanding the problem you're trying to solve is the key that unlocks the door to solutions to a problem. So, whenever I am faced with a challenge, I make sure that I sufficiently understand the problem at hand, research more about the problem, and review the literature to see how others have tried to solve the same or related problems, before starting to tackle the problem.

When I feel that I have sufficiently understood the problem and possible approaches to solving it, I then start trying out different approaches learnt. Most importantly, if provided, the starter notebook provides a very important starting point!

Words of encouragement for others, or advice that has helped you?

They say, "If you want to go fast, go alone. If you want to go far, go with others!" This has proven to be true. With collective efforts as a team, you go far as a result of complementing each other's efforts. I encourage team work!

What do you like about Zindi?

It is a good place for growth. Continuous improvement of the platform is making it more and more resourceful. Most importantly, it is user and beginner friendly.

What would you like to see more of on the platform?

I want to see winner badges for every challenge won!

Back to top
If you enjoyed this content upvote this article to show your support
Discussion 0 answers