We held another professional breakfast, this time on the topic of credit forecasting. The breakfast was mainly directed at professionals from the fields of AI, data analysis, and modelling, but we also gladly welcomed directors and managers, especially from the banking sector as they are directly affected by the topics from a business perspective. After all, the highlight of the breakfast was a presentation of the results of our collaborative research on credit forecasting, done with the risk department of Equa Bank a.s.

With our partners from Equa Bank, we built a propensity model to identify clients who would apply for loan products in the future. The model combines the knowledge we gathered from our research on pseudo-social networks with modern machine learning techniques. The results presented by the director of credit risk at Equa Bank, Mr. Peter Baláži, and data scientist, Mr. Martin Hazal, greatly exceeded our expectations. The comparison between forecasted and actual loans during the specified time period clearly showed very good usability of the model at approximately 70% of all bank clients, including less active segments!

Then Petr Paščenko, the leader of our data science team, showed us how the propensity model can be used to address clients and in marketing campaigns. Above all, however, he explained how it is possible to directly assess the financial benefits of deploying the propensity model in a bank’s environment. A treat for specialists in the field was the look under the hood at the actual operation of the model, including a completely innovative application of graph embedding for the extraction of flags from the transaction data.

In the second block, our colleague Petr Hála talked about our research on modelling social ties. We already did some testing on Friday on the model for detecting clients who live in the same household. Now we have added the anonymous detection of friends and people who clients spend time with to our solution. The lecture described how the right application of a simple statistical concept can help read social ties from transactions between clients and the bank.

At the very end, our DevOps specialist, Tomáš Karella, talked about his paper on the development of models and their deployment on the cloud. This paper was very interesting in that not only did it go over the advantages and disadvantages of processing data on the cloud but, above all, it simply and very clearly described what is behind the recently widely-used designation MLOps.

We are very pleased that our data science breakfast was a great success, and we are looking forward to organizing more successful events in the future!

Author: Lukáš Dvořák

Big Data & Data Science Presales Evangelist