Assessing Financial Well-Being of Merchants by Analyzing Behavioral Patterns in Historical Transactions

Assessing Financial Well-Being of Merchants by Analyzing Behavioral Patterns in Historical Transactions

Kumar Abhinav Srivastava (Massachusetts Institute of Technology, USA), Vivek Kumar Singh (Rutgers, The State University of New Jersey, USA & Massachusetts Institute of Technology, USA), Burcin Bozkaya (Sabanci University, Turkey) and Alex “Sandy” Pentland (Massachusetts Institute of Technology, USA)
Copyright: © 2015 |Pages: 18
DOI: 10.4018/978-1-4666-8465-2.ch003
OnDemand PDF Download:
No Current Special Offers


This study focuses on a new approach to estimate financial wellbeing indicators for merchants, by looking at behavioral patterns of their customers using transaction history. The transaction data for about 10,000 merchants in a specific country, was analyzed in terms of diversity and propensity of their customers using factors like age, distance travelled to shop, time of the day for shopping, day of the week for shopping, educational status, gender etc. These factors were used as independent variables to predict the financial well-being of merchants, particularly in two dimensions – total revenue and consistency in revenue, both relative to other merchants in the same industry. The merchants were then also divided into the categories of Essential, Non- essential and Luxury goods depending on the industry they belong to and it was interesting to observe the contrast across categories. The results suggest that behavioral patterns could be used to augment current methods of calculating credit score.
Chapter Preview


There has been a lot of research in estimation of Credit risk using machine learning and artificial intelligence. Credit rating analysis was initially done using statistical methods. Later, modeling using Artificial Intelligence showed better results over time. One research paper by Huang et al (2003) was focused on comparing explanatory power of Back-propagation Neural Network (BNN) and Support Vector Machines (SVM) using the data from US and Taiwan markets. It suggested that SVM gives comparable if not better results than BNN.

Another study by Huang et al. (2007) showed that SVM model for Credit Card scoring can achieve similar results as neural networks, genetic programming, and decision tree classifiers while using lesser number of inputs. Comparison of performance with multiple algorithms like Back Propagation, Extreme Learning Machine, Incremental Extreme Learning Machine and Support Vector Machine, was the focus of the research work by Zhong et al. (2013).

Complete Chapter List

Search this Book: