Innovation in customer scoring for the financial services industry
- PDF / 110,630 Bytes
- 2 Pages / 439.37 x 666.142 pts Page_size
- 63 Downloads / 189 Views
Innovation in customer scoring for the financial services industry Arno De Caigny © Springer-Verlag GmbH Germany, part of Springer Nature 2019
This is a summary of the author’s Ph.D. thesis supervised by Kristof Coussement and Koen W. De Bock, and defended on August 26th, 2019 at Université de Lille. The thesis is written in English and is available from the author upon request at [email protected]. The author is affiliated with IESEG School of Management and LEM-CNRS 9221, both located at 3, rue de la Digue, 59000, Lille, France. The author extends his gratitude to Crédit Agricole Nord de France for financial and logistic support during his Ph.D. This work deals with customer analytics, which is a vital aspect in effective customer relationship management. In recent years, analytics and data science have gained considerable attention in the operations research community. Multiple contributions in three areas of innovation in customer scoring are pursued: methodology, data and application. On the methodological level, there are three notable contributions to the innovative applications in the operations research stream. First, in many customer scoring applications not only predictive performance but also comprehensibility is important. Therefore, a new algorithm, named the logit leaf model, is presented that combines decision trees and logistic regression to fulfil both predictive performance and comprehensibility requirements. Moreover, it automatically creates segments which allows analysts to detect segment-specific drivers. Such drivers can provide valuable insights as demonstrated in a case study. The logit leaf model is benchmarked in a customer churn prediction context on 13 real-world data sets against four conceptually related and popular algorithms. The results show superior performance of the logit leaf model over its building blocks, i.e. decisions trees and logistic regression, and the logit leaf model performs at least as well as the more complex ensemble methods random forest and logistic model trees. Second, improvements to methodology for innovative data sources are presented. A convolutional neural network has been used to process textual data in order to incorporate it in a customer churn prediction model. The results on a real-world customer churn prediction case show that this deep learning based methodology performed significantly better than the current state-of-the-art vector space-based approach. This dissertation also deals with Big Data as it extends methodology to incorporate fine-grained data in a customer scoring model. In order to maintain the specifics of the fine-grained data as much as possible, a pseudo-social network is created based on similarities in payment behavior. From this network based on Big Data, behavioral similarity scores are derived. Existing methodology for such
123
A. De Caigny
payment data include only whether or not people share certain counterparties while in our methodology also the recency, frequency and monetary value are used to create richer similar
Data Loading...
