Title: Adding value to credit scoring using mobile phone data and social networks
Authors: Maria Oskarsdottir - KU Leuven (Belgium) [presenting]
Abstract: One of the oldest applications in analytics is credit scoring where, traditionally, peoples banking history is used to assess their creditworthiness. However, as data is continuously being generated in more volume and variety than ever before, new credit assessment methods are emerging. In particular, new variables to capture borrower behavior going beyond simple repayment history have been shown to be good predictors of whether or not people will default on their loans. We show how both statistical and economic model performance of credit scoring models can be enhanced by incorporating alternative data sources. We build networks using call-detail records and extract features that represent the calling behavior of customers. In addition, we apply social network analytics techniques where we propagate influence from prior defaulters throughout the network to inspect the influence of known defaulters on others. The resulting influence features and the calling behavior features are used together with traditional bank history features when building the scorecards. According to our results, including network information in credit scoring models significantly increases their performance when measured in AUC. Furthermore, we demonstrate benefit in terms of profit by applying a profit measure, which we further expand for profit-based feature selection.