A two‐step system for direct bank telemarketing outcome classification

• Published date: 01 January 2017
• Author: Salim Lahmiri
• DOI: http://doi.org/10.1002/isaf.1403
• Date: 01 January 2017

RESEARCH ARTICLE

A two‐step system for direct bank telemarketing outcome

classification

Salim Lahmiri

ESCA School of Management, Casablanca,

Morocco

Correspondence

Salim Lahmiri, ESCA School of Management,

Casablanca, Morocco.

Email: slahmiri@esca.ma

Summary

A two‐step system is presented to improve prediction of telemarketing outcomes and to help the

marketing management team effectively manage customer relationships in the banking industry.

In the first step, several neural networks are trained with different categories of information to

make initial predictions. In the second step, all initial predictions are combined by a single neural

network to make a final prediction. Particle swarm optimization is employed to optimize the initial

weights of each neural network in the ensemble system. Empirical results indicate that the two‐

step system presented performs better than all its individual components. In addition, the two‐

step system outperforms a baseline one where all categories of marketing information are used

to train a single neural network. As a neural networks ensemble model, the proposed two‐step

system is robust to noisy and nonlinear data, easy to interpret, suitable for large and heteroge-

neous marketing databases, fast and easy to implement.

KEYWORDS

classification,neural networks, particle swarmoptimization, subsystems, telemarketing outcome

1|INTRODUCTION

Telemarketing is a direct marketing used to provide accurate solu-

tions for the brand owner, including marketing products and services,

and finding potential clients. As a cost‐effective process between the

company and its customers, telemarketing is rapidly growing around

the world within a variety of industries, including telecommunica-

tions, banking, insurance and other financial services. There is an

abundant literature in risk and business failure prediction as it is an

important issue in banking risk management (Abdou & Poiton,

2011; Çelik, 2013; Figini, Savona, & Vezzoli, 2016; Li, Yu, Zhou, &

Cai, 2013; Peat & Jones, 2012; Pendharkar, 2011; Quek, Zhou, &

Lee, 2009; Savona & Vezzoli, 2012; Sun, 2012; Trinkle & Baldwin,

2007). However, little attention has been given to the problem of

predicting the success of telemarketing calls for selling bank long‐

term deposits (Javaheri, Sepehri, & Teimourpour, 2014; Lau, Chow,

& Liu, 2004; Moro, Cortez, & Rita, 2014; Moro, Laureano, & Cortez,

2012). Indeed, marketing data mining and analytics are essential to

companies to choose the right strategic decision‐making policies for

selling the right product to the right customer to better generate

profits (You et al., 2015), to acquire new customers and increase cus-

tomer loyalty (Duman, Ekinci, & Tanrıverdi, 2012), and to analyse

attitudes and emotions of the customer, and the potential factors

that influence them (Montañés, Suárez‐Vázquez, & Quevedo, 2014).

Therefore, the development of an effective system to predict the

outcome of a telemarketing call is a very important issue to choose

the right selling strategy and to increase customer loyalty. In partic-

ular, such a predictive system is expected to be effective in model-

ling large and heterogeneous marketing databases, to be accurate,

and to be easy to interpret.

Indeed, in the context of bank telemarketing success prediction,

Lau et al. (2004) focused on the issues appropriate to the database

approach to cross‐selling in the banking industry and explained the

potential of application of data mining methods in bank telemarket-

ing. However, no classification or prediction model was developed

and tested. Moro et al. (2012) examined the effect of attributes on

accuracy of a support vector machine (SVM) classifier in the predic-

tion of deposits subscription based on direct marketing campaigns. In

other words, the purpose was to conduct a sensitivity analysis. They

concluded that while they provided interesting information for bank

telemarketing campaign managers, their model cannot be used for

prediction purposes. Javaheri et al. (2014) analysed how a mass

media marketing campaign could affect the buying of a new bank

product. The SVM classifier achieved 81% classification accuracy

when trained with 26 attributes selected using F‐statistics. In a

recent study, Moro et al. (2014) employed artificial neural networks

Received: 1 July 2016 Revised: 30 November 2016 Accepted: 14 January 2017

DOI 10.1002/isaf.1403

Intell Sys Acc Fin Mgmt. 2017;24:49–55. Copyright © 2017 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/isaf 49