A divide‐and‐conquer strategy using feature relevance and
expert knowledge for enhancing a data mining approach to
Instituto Universitário de Lisboa (ISCTE‐IUL),
ISTAR‐IUL, Lisboa, Portugal
ALGORITMI Research Centre, University of
Minho, Guimarães, Portugal
CIS‐IUL, Instituto Universitário de Lisboa
(ISCTE‐IUL), Lisbon, Portugal
NOVA Information Management School
(NOVA IMS), Universidade Nova de Lisboa,
Sérgio Moro, Instituto Universitário de Lisboa
(ISCTE‐IUL), ISTAR‐IUL, Lisboa, Portugal.
The discovery of knowledge through data mining provides a valuable asset for addressing deci-
sion making problems. Although a list of features may characterize a problem, it is often the case
that a subset of those features may influence more a certain group of events constituting a sub‐
problem within the original problem. We propose a divide‐and‐conquer strategy for data mining
using both the data‐based sensitivity analysis for extracting feature relevance and expert evalua-
tion for splitting the problem of characterizing telemarketing contacts to sell bank deposits. As a
result, the call direction (inbound/outbound) was considered the most suitable candidate feature.
The inbound telemarketing sub‐problem re‐evaluation led to a large increase in targeting perfor-
mance, confirming the benefits of such approach and considering the importance of telemarket-
ing for business, in particular in bank marketing.
banking, data mining, divide and conquer, feature selection, marketing
Data mining (DM) enables to unveil previously undiscovered knowledge, providing leverage for decision making (Witten, Frank, Hall, & Pal, 2016).
The basic ingredient for DM is raw data, representing known instances of an interesting phenomenon to be modelled (i.e., the problem) that can be
characterized by a list of features. A key aspect for a successful DM project is feature engineering, because data features need to be related with
the problem modelled (e.g., desired outcome to predict; Wang, Chen, & Bi, 2015). In previous works, we have addressed the telemarketing problem
of predicting the result from phone call campaigns to sell bank deposits using DM. First, a bank telemarketing dataset was enriched with social and
economic context features, leading to a tuned model that enabled to reach 79% of the deposit subscribers by selecting the half better classified
clients (Moro, Cortez, & Rita, 2014). Such a model was then improved by including customer lifetime‐value‐related features, increasing the perfor-
mance to 83% of subscribers with the half better classified contacts (Moro, Cortez, & Rita, 2015b).
In this paper, we further enhance the classification model by presenting a divide‐and‐conquer strategy for splitting the problem of selecting
the most likely subscribers of a bank long‐term deposit offered in the context of telemarketing campaigns, where contacts were conducted
through phone calls within the bank's contact centre. The proposed approach is focused on selecting a candidate feature among the input
features used to feed the model for logically dividing the problem into sub‐problems. Then, a specific sub‐problem is assessed for unveiling
the benefits of the new classifier focused on a particular context within the larger problem. In this article, the next section describes the telemar-
keting and DM background related with this paper, followed by a section to present the bank dataset, the methods used for the divide‐and‐
conquer strategy proposed and for assessing the impact on the new sub‐problem. Then, the results are discussed. Finally, conclusions are
summarized in the last section.
Received: 13 September 2016 Revised: 19 September 2017 Accepted: 22 September 2017
Expert Systems. 2018;35:e12253.
Copyright © 2017 John Wiley & Sons, Ltdwileyonlinelibrary.com/journal/exsy 1of13