Adacta and Raiffeisen bank in a Big Data student competition
Mozgalo, the first and the only student competition in the deep analysis of large amounts of data and the creation of a Big Data solutions took place in Zagreb, Croatia.
The competition is organised every year by the student association eSTUDENT. This year Adacta and Raiffeisen Bank Croatia created the task for students participating in the competition. The task was to predict premature loan repayment using data science techniques to develop client behaviour machine-learning models.
Client behaviour modelling
The machine-learning tasks targeting loan repayment are usually focused on recognising clients who may not be able to repay the loan. The Mozgalo competition task came with a slight twist, the students had to recognise, and predict the clients that would return the loan, or withdraw the deposit prematurely.
Adacta and Raiffeisen Bank workshop for students
Through the three months of the competition before preparing their final presentations, students had a chance to participate in workshops held by partner companies. Božidara Cvetković (Adacta) together with Irena Kovačević (Raiffeisen Bank Croatia), Ivana Jelas (Raiffeisen Bank Croatia) and Gordana Jakšetić (Raiffeisen Bank Croatia) provided valuable insights and shared their knowledge with students to prepare them for the challenges they were going to face until the finals of this year’s competition.
Machine learning methodology and the usage of real and noisy data
In an introductory workshop, Raiffeisen Bank experts introduced the problem of premature loan repayment and premature deposit withdrawals, while Adacta introduced the dataset and the methodology for solving the problem using data analysis and machine-learning for client behaviour modelling. The workshop covered the essentials in all the required steps of the pipeline from data fusion and data cleaning to training a machine-learning algorithm to output the predictions.
In addition to client behaviour modelling, the focus of this task was on the utilisation of third-party data (in this case the economic indicators) to enrich the client data and improve the understanding of the problematic events.
Machine learning is usually taught on synthetic or clean datasets, here the students were able to try the introduced methodology using real and noisy data. The entire procedure was set in such a manner as to experience the obstacles that may arise when you try to tackle a real-life problem with machine-learning.
Adacta and RBA support throughout the entire competition
To simplify the submission and evaluation process, the online evaluation of the solutions was set-up. Each team was able to submit their solution multiple times a day and see how they compare with other teams. The questions were collected by the eSTUDENT team and sent to Adacta once per week, which was then able to provide answers to the student’s concerns.
The final results presented to data science experts
After three months of continuous work and study on the competition task, the teams submitted their solution presentations. Six teams made their way into the final stage which was held this June in Zagreb. The finalists presented their solutions before the jury of Data Science experts from the partner companies.
The jury of Data science experts during student presentations
“All the solutions that were selected for the finals were interesting, and it was a pleasure to hear them present their work. We could see solutions using the single machine-learning model, solutions that used ensembles and the winning solutions that used stacking. I really liked the data analysis of the economic indicator in relation to the dataset and the creativity in constructing a single or multiple-model final solution.”
Božidara Cvetković, Data Scientist, Adacta
Valuable experience with real data and machine learning
The BiotechRi team from Rijeka took the first prize while Pessoe and Pet matematičara teams took second and third place respectively. Besides the prizes prepared by the partner companies and money funds, this competition gave a lot more to the students. They gained experience with real data and knowledge on client behaviour prediction that will help them at the start of their professional careers. Adacta is very proud to have had a chance to be part of this competition and to contribute to the student community with their knowledge and support.