Hsin Vonn Seow

Benchmarking state-of-the-art classification algorithms for credit scoring: An update of research

Many years have passed since Baesens et al. published their benchmarking study of classification algorithms in credit scoring [Baesens, B., Van Gestel, T., Viaene, S., Stepanova, M., Suykens, J., & Vanthienen, J. (2003). Benchmarking state-of-the-art classification algorithms for credit scoring. Journal of the Operational Research Society, 54(6), 627–635.]. The interest in prediction methods for scorecard development is unbroken. However, there have been several advancements including novel learning methods, performance measures and techniques to reliably compare different classifiers, which the credit scoring literature does not reflect. To close these research gaps, we update the study of Baesens et al. and compare several novel classification algorithms to the state-of-the-art in credit scoring. In addition, we examine the extent to which the assessment of alternative scorecards differs across established and novel indicators of predictive accuracy. Finally, we explore whether more accurate classifiers are managerial meaningful. Our study provides valuable insight for professionals and academics in credit scoring. It helps practitioners to stay abreast of technical advancements in predictive modeling. From an academic point of view, the study provides an independent assessment of recent scoring methods and offers a new baseline to which future approaches can be compared.

Using adaptive learning in credit scoring to estimate take-up probability distribution

Abstract Credit scoring is used by lenders to minimise the chance of taking an unprofitable account with the overall objective of maximising profit. Profit is generated when a good customer accepts an offer from the organisation. So it is also necessary to get the customers to accept the offer. A lender can “learn” about the customers’ preferences by looking at which type of product different types of customers accepted and hence has to decide what offer to make. In this model of the acceptance problem, we model the lenders decision problem on which offer to make as a Markov Decision Process under uncertainty. The aim of this paper is to develop a model of adaptive dynamic programming where Bayesian updating methods are employed to better estimate a take-up probability distribution. The significance of Bayesian updating in this model is that it allows previous responses to be included in the decision process. This means one uses learning of the previous responses to aid in selecting offers best to be offered to prospective customers that ensure take-up.

To ask or not to ask, that is the question

Abstract Applicants for credit have to provide information for the risk assessment process. In the current conditions of a saturated consumer lending market, and hence falling “take” rates, can such information be used to assess the probability of a customer accepting the offer? With the advent of internet broking pages, which allow borrowers to “apply” to a number of different companies at the same time, this “take” problem will increase. In some mortgage markets, it is quite common for more than 50% of those offered credit to reject it. In some cases, this is because the sale falls through but often it is because a relatively better retailer has offered a more suitable product to the borrower. Lenders do not want to make the application process too complicated, and with the growth in adaptive marketing channels like the Internet and the telephone, they can make the questions they ask depend on the previous answers. We investigate how one could develop such “adaptive” application forms; which would assess acceptance probabilities.

Using a transactor/revolver scorecard to make credit and pricing decisions

In consumer lending the traditional approach is to develop a credit scorecard which ranks borrowers according to their risk of defaulting. Bads have a high risk of default and Goods have a low risk. To maximise the profitability of credit card customers, a second classification between revolvers and transactors becomes important. Building a transactor/revolver scorecard together with a Good/Bad scorecard over the revolvers, gives rise to a risk decision system whose ranking of risk is comparable with the standard approach. The paper develops a profitability model of card users including the transactor/revolver score leads. This gives more accurate profitability estimates than models which ignore the transactor/revolver split.

Modified TAROT for cross-selling personal financial products

The Top Application characteristics Remainder Offer characteristics Tree (TAROT) was first introduced in 2007. This is a modified Classification and Regression Trees (CART) used to help decide which question(s) to ask potential applicants to customise an offer of a personal financial product so that it would have a high probability of take up. In this piece of work the authors are presenting, they have further modified the TAROT to cross TAROT, using its properties and modeling steps to deal with the issue of cross-selling. Since the bank already has ready customers, it would be ideal to cross-sell the financial products seeing that one can ask one (or more) further question(s) based on the initial offer to identify and customise another financial product to offer.

An improved GRMOD heuristic for container loading problem

The Container Loading Problem (CLP) is a study of loading a subset of goods or parcels of different sizes into a three-dimensional rectangular container of fixed dimensions such that the volume of packed boxes is maximized. In this paper, an improved version of the modified George and Robinson heuristic (iGRMOD) is developed to solve the CLP. Comparison computational results on benchmark data set from the literature will be presented. The performances of the iGRMOD are superior than the GRMOD and other heuristics reported in the literature.

ROUTING VEHICLES WITH ANTS

Routing vehicles involve the design of an optimal set of routes for a fleet of vehicles to serve a number of customers with known demands. This research develops an Ant Colony Optimization for the vehicle routing with one central depot and identical vehicles. The procedure simulates the behavior of real ants that always find the shortest path between their nest and a food source through a form of communication, pheromone trail. Finally, preliminary results on the learning of the algorithm testing on benchmark data set will be presented in this paper.

LOADING CONTAINERS WITH ANTS

Loading containers is like loading a subset of given three-dimensional rectangular boxes of different sizes into a three-dimensional rectangular container of fixed dimensions in order to achieve optimal space utilization. In this paper, Ant Colony Optimization (ACO) with its probabilistic decision rule is used to construct towers of boxes and to arrange them into the container. Some initial computational results on benchmark data set will be presented.

Question selection responding to information on customers from heterogeneous populations to select offers that maximize expected profit

The advent of Internet broking pages allows customers to ‘apply’ to a number of different companies at one time, leading to multiple offers made to a customer. The saturated condition of the personal financial products has led to falling ‘take’ rates. Financial institutions are trying to increase the ‘take’ rates of their personal financial products. Applicants for credit will have to provide information for risk assessment, which can be used to assess the probability of a customer accepting an offer. Interactive channels such as the Internet and telephone allow questions that are asked to depend on previous answers. The questions selected need to provide information to assess the probability of acceptance of a particular variant of financial product. In this paper, we investigate a model to predict the best offer to extend next to a customer based on the response for the questions, as well as the question selection itself.