The credit scoring industry has a long tradition of using classical statistical methods for both short- and long-term loan default probability predictions. These methods use domain specific variables and standards that have been established long before the hype of machine learning. Although several commercial software companies offer specific solutions for credit scorecard modelling, their methods are not explicitly clear. The basic difference of traditional modelling and machine learning is that “in traditional modelling one intends to set up a modelling framework and try to establish relationships while in machine learning we allow the model to learn from the data by understanding the hidden patterns”.
Hence the first one is subjective and thus prone to bias as it requires analyst to have solid understanding of statistical techniques and business knowledge while the later one is more complex in nature and computationally intensive, hence requires higher computation power of the systems and analyst needs to be tech savvy. Our view is that both approaches should be used, and adjustments made as appropriate to credit decisions depending on inferences drawn from each method.
Contact Us