Meet the Winners of the Hulkshare Recommendation Algorithm Challenge
Meet the winners · 8 Sep 2022, 09:34 · 4 mins read ·

Meet Cobus Burger and Adeyinka Michael Sotunde, winners of the Hulkshare Recommendation Algorithm Challenge as they share their experience and what gave them the winning edge.

In the Hulkshare Recommendation Algorithm Challenge, Zindians were tasked with developing a machine learning algorithm to predict whether someone enjoyed a song on the streaming service Hulkshare using the frames of a song, duration of the song and the viewing pattern.

Read on to find out what set these winners’ solutions apart from the rest.

Winner, Cobus Burger, South Africa.

Please introduce yourself.

My name is Cobus Burger aka cobusburger. I work as a Development Economist at Stellenbosch University, and as a Senior Data Scientist at Predictive Insights.

Please explain your solution and what set your winning solution apart from others.

I had a single ordinary least square (OLS) and a single light GB model that did similarly well. I took a weighted average of the two for my final submission.

Most of the explanatory power of my model came from how many times a song had been listened to and who listened to it. If someone listened to 5 songs in the training data and we can see that these songs were popular then we know this person has a broad taste and that the songs they listened to in the test data will probably also be popular. I believe what had me winning was using fixed effects - I exploited the fact we had songs from the same session in both the train and test data and used ordinary least square (OLS) models.

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

You can't rush feature engineering. I usually start a competition, do the basics, and then leave it for a week just so I can revisit it with a fresh perspective

What do you like about Zindi?

I love that we all have the same data and that we are watching the scoreboard together.

How do you prepare for a challenge?

I always start by drawing to show the relationship between each variable and the outcome so I can get a feel for each variable and how important it is. This guides me during feature engineering.

5th-placed Goal-Oriented-Team, represented by Adeyinka Michael Sotunde, Nigeria.

Please introduce yourself.

I am Adeyinka Michael Sotunde aka Micadee (Goal-Oriented-Team) from Lagos, Nigeria. I am a Data Scientist, Kaggle Expert and currently working as the Senior Budget/Planning Officer and Data Scientist at Lagos State House of Assembly.

Please explain your solution and what set your winning solution apart from others.

Our approach was based on creating some statistics or statistical features from the given datasets, as well as implementing the rolling mean to capture vital information and insights from the datasets provided. Apart from having a great set of ideas from feature engineering, and considering the large datasets provided, the major difference is we were able to apply a package called pandarallel to our extraction notebooks and this sped up the running time to just a few seconds. This inspired us to further explore more datasets.

What are the biggest areas of opportunity you see in AI in Africa over the next few years?

The rapidly developing set of artificial intelligence (AI) technologies has the potential to solve some of the most pressing challenges that impact Sub-Saharan Africa and drive growth and development in core sectors:

  • Agriculture will be done more efficiently and effectively, raising yields
  • Healthcare will be better tailored, higher quality, and more accessible, improving outcomes
  • Public services will be more efficient and more responsive to citizens, enhancing the impact
  • Financial services will be more secure and reach more citizens who need them, expanding access

What are you looking forward to most about the Zindi community?

I am happy with the current rapid progress and growth of the Zindi community. I look forward to more corporate sponsors coming in and thereby leading to solving most of the major problems facing Africa.

What do you like about Zindi?

I like the current upgrade of the Zindi platform interface, and the launch of a recruitment platform to connect organisations with Africa’s data science talent. I must say that Zindi is doing a great job. Cheers!

What is one thing that the Zindi platform does well?

Zindi has impacted a lot of the young generation, as it serves as a platform to practise skills and learn some new skills as well.

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