InData Labs at Imaguru Data Challenge, Datathon 2019
Two weeks back, 19-21 of July, our team took part in the 3rd Datathon, Imaguru Practical Data Challenge. The competition was held with the support of the US Agency for International Development, USAID, in Belarus.
Over the three days, 12 teams consisting of 120 participants worked hard to nail the tasks by Datathon partner companies: Menu.by, SP&P, and BNB-Bank. Real business challenges required innovative and accurate solutions driven by data science and big data techniques.
The program also included a Design Thinking course — a phase of brainstorming ideas and cooperating with data science mentors from various industries ready to help come up with practical ways of solving the tasks. The result of Datathon was a generated team around the tasks and developed prototype using a provided dataset.
For the third time, InData Labs was invited as an exclusive data science partner. Our experts provided participants with mentoring support and also delivered masterclasses on machine learning.
Welcome our Mentors!
Anton Kulesh, Data Scientist at InData Labs. In his report “Explaining predictions of machine learning models”, Anton referred to the issue of interpreting “black boxes”, that is, the models used to solve a wide range of applied tasks. He dwelled upon the importance of this topic, outlined the existing methods and tools and the ways of using them within supervised learning (e.g., customer churn prediction). Besides, he highlighted some strengths and weaknesses of the approaches.
Alexei Tishurov, Machine Learning Engineer at InData Labs, delivered a masterclass “Task formulation, validation, baselines, and other boring yet important stuff”. He explained why it is vital to define clearly the tasks of machine learning, choose the right methods to validate the model quality and have baseline solutions to compare.
Results & Winners
During the event, all participants could work either with the datasets provided by partner companies or some open-source datasets most suitable for solving the tasks.
Out of all the teams that used the dataset from Menu.by, the winner was BP Team. They worked on a recommendation system able to predict user preferences and, based on that, give recommendations. The recommendations were generated by collaborative filtering using SVD decomposition.
BNB-Bank chose A7 forecast as the winner. The team worked on creating a model to forecast the loan life cycle.
After processing and analyzing data on more than 3,700 different loans, they identified 4 segments according to the types of loans, characteristics of recipients, and loan histories. Based on the obtained data, the team presented the model capable of classifying a new client into 1 of the 4 segments and predict loan repayment.
Of the teams working with open-source datasets, the winner was Friday team. Their solution was a database of documents on antitrust issues for SP&P law firm.
The team created an application to enable a quick search of documents based on multiple parameters. They also highlighted 5 ways to improve database search using machine learning methods.
InData Labs would like to thank Imaguru and all partners of Datathon for organizing such a great event!
We congratulate the winners and send our best wishes to all participants!