13 Dec 2019, 11:37

Meet the winners of the Xente Credit Scoring Challenge

Zindi is excited to announce the winners of the Xente Credit Scoring Challenge! The challenge attracted 335 data scientists from across the continent and around the world, of whom 96 were on the leaderboard. Some of our winners share their solutions.

Congratulations to DrFad, Team Hi (Blenz, FADHLOUN, and Mohamed_Salam_Jedidi), and Team Dom_2 (Engineer, Lawrence_Moruye, Warrie_Warrie_dsn, and OLALEYE_ENIOLA_DSN).

In this challenge, Zindi users aimed to create a machine learning model to predict which Xente customers are most likely to default on their loans, based on their loan repayment behaviour and ecommerce transaction activity. Xente is a Ugandan e-commerce startup that makes it easy for consumers to make payments, get loans, and shop using a mobile phone.

The challenge attracted 335 data scientists from across the continent and around the world, of whom 96 were on the leaderboard. We are happy to introduce representatives of the top two teams to share their approach to this challenge: Mohammed Amine and Youssef Fadloun (Tunisia, 2nd place), and Adegunle Ahmed Babatunde (Nigeria, 3rd place).

Benlamine Mohamed Amine

Zindi handle: Blenz

Where are you from? Tunisia

Tell us a bit about yourself:

I'm a data scientist at Neopolis Development, we're based in Nabeul,Tunisia. I started my data science journey 10 months ago, and now I'm lucky to be able to compete on Zindi and learn from other people.

Tell us a bit about the approach you took:

I took my time understanding the problem, the data, and the target to be solved, then started extracting domain knowledge features such as counts on missed payments; mean, min, and max amount spent by every customer in his transactional history, etc. I also engineered some features during loans and outside loan time-frames in order to detect any changes in behavior on the customer's part. The data was small, so I used a single catboost classifier with manual tuning and feature selection.

What were the things that made the difference for you that others can learn from?

Taking the time to build intuition behind the data patterns. If you can express those patterns in layman's terms, you can simply play around with features later on to find insights.

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

I think the AI disruption will affect every area in Africa, from transport to education.

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

I'm just hoping the community gets bigger, more people get interested in competing with us, and we can eventually benefit each other with different ideas.
Link to Github Repo.

Youssef Fadhloun

Zindi handle: FADHLOUN

Where are you from? Tunisia

Tell us about yourself:

I’m a data scientist and mathematician in a wide range of functions including predictive modeling, content discovery, product analytics and programming skills, particularly Python.

What were the things that made the difference for you that others can learn from?

I think that Exploratory Data Analysis is the most important step in a data science pipeline that most of the competitors neglect - without it it's almost impossible to unravel the secrets of the data that will affect the rest of the pipeline. Sure you can model and get a better score than others, but you can never shine .

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

I think the correlation coefficient between medicine and AI will increase in the coming years. I hope we can steer this new technology to save the lives of people dying in Africa and also detect diseases like cancer at its earliest stages.

Adegunle Ahmed Babatunde

Zindi handle: Engineer

Where are you from? Ogun State, Nigeria

Tell us about yourself:

I’m a recent graduate of Ekiti State University, where I studied Civil Engineering. I like to make data speak to solve problems and revolutionise industries for good.

Tell us a bit about your the approach you took:

My pipeline was as follows: Hypothesis generation - Exploratory data analysis - Data preprocessing - Feature Engineering/Extraction - Modelling. Then I iterated this pipeline to make the model better.

What were the things that made the difference for you that others can learn from?

I took my time to visualise the data by plotting both univariate and bivariate features. Also doing the right preprocessing and selecting the right features for my model helped a lot.

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

Clustering analysis and Natural language preprocessing are going to be a big fish in Africa.

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

To learn from the best of the best and challenge the best of the best for the greater good.
Link to Github Repo.

Zindi would like to thank the challenge winners, everyone who participated in the challenge, as well as competition hosts Xente and Standard Bank.