As the largest bank in Norway, DNB collects huge volumes of customer data. This gives the bank good prospects for utilizing both traditional statistical models and new machine learning algorithms. The department in which the candidate is working is one of the bank’s leading environments within customer analytics. An important part of the departments responsibilities is running a “Model Factory”, which is a framework for development of prediction models. Most of these models are used to predict which customers are interested in different products, thereby enabling personalized communication. The factory has been used for approximately seven years. During this time, there has been an immense technological development within machine learning and artificial intelligence, both in terms of new algorithms and more processing power. This makes way for implementing more flexibility in the framework in order to fully utilize the new opportunities. DNB therefore wants to support the candidate in exploring different methods for finding the optimal settings when testing different models. The starting point of the exploration will be design of experiments, as it was the theme of the candidate's Master thesis. Combining design of experiments with machine learning is a promising, but mostly unexplored terrain. Successful implementation will increase interpretability and thus responsible implementation of the models. Different aspects of modelling can be included in the design in order to incorporate relevant challenges from the Model Factory. The results can then be used to guide further work. Better models will be beneficial for the bank as the customer contact can be made even more targeted, potentially resulting in more sales and more satisfied customers. Successful recommendations will also be shared with other relevant analytical departments of the bank. Of particular interest is how to handle unbalanced data, as this is an issue faced by many of the modelling departments.
Project leader: Jacob Kooter Laading
Institution: DNB ASA