direct marketing in banking using data mining

On Identifying Potential Direct Marketing Consumers
using Adaptive Boosted Support Vector Machine
Armin Lawi Department of
Computer Science Universitas
Hasanuddin
Makassar, Indonesia
armin@unhas.ac.id
Ali Akbar Velayaty*)
Post-graduate Program of Electrical
Engineering, Universitas Hasanuddin
Makassar, Indonesia
arvasit@gmail.com
Zahir Zainuddin
Dept. Electrical Engineering
Universitas Hasanuddin
Makassar, Indonesia
zainuddinzahir@gmail.com
Abstract— Identifying potential consumers for direct
marketing to very large data is a difficult and impossible
task to do manually. Therefore, the machine learning
approach needs to help analyze the data to contribute in
determining the marketing strategy policy. In this paper,
vector machine support methods using the Adaboost
algorithm are investigated to classify potential customers
for direct marketing of large bank data marketing. The
adaboost algorithm aims to build a better model than the
model generated from a single classifier. Data obtained
from UCI machine learning repository. Total data is
processed as many as 9280 data with the number of
attributes of 20 classes and 1 target class. Training data and
test data are divided into 70% and 30%. This classification
predicts the prospects for a deposit subscription. The results
show that the SVM method using Adaboost algorithm
obtained accuracy is 95.07% and the sensitivity is 91.65%
higher than the ordinary SVM approach.
Keywords—Bank Direct Marketing; Support Vector Machine;
Adaboost.
I. INTRODUCTION
Business Intelligence is a term related to the use of
information space and Intelligence mechanisms to support a
decision [1]. One sector that implements this is the banking
sector that receives and processes information from customers
every day. One area of banking that collects information from
customers is the marketing sector. There are two ways of
banking in introduce the goods or services that is with the
general advertising either through television, radio, newspapers,
etc. or by targeting customers in particular or called the bank
direct marketing [2].
Direct marketing is a marketing technique applied to the
bank to introduce a product either in the form of goods or
services to customers by phone, email and others [3]–[5].
However, the disadvantages of direct marketing are those
customers who feel disturbed that can cause negative ratings for
the bank and the cost and time spent by telemarketers is not
proportional to the number of customers contacted [2]
Over time, the amount of data that contains information from
customers will increase. The increasing amount of data can be
used to find information that serves to assist in making a
decision. One of the computational methods that discuss the
above problem is Data Mining [1].
Data mining is a field in the computational process that
allows finding a pattern of a data that is often ignored by using
a particular method. The pattern can be used to help take in a
decision. One method of data mining is a classification method
that is a technique of grouping data obtained based on the pattern
produced previously. There are several models of classification
among them is Support Vector Machine (SVM) method.
Support Vector Machine is the method that forms the best
separator in hyperplane field on the data that separates the two
classes by maximizing the margin value of the two classes.
According to [6] Although ordinary SVM shows good
performance in classification, but is sensitive to sample and
parameter settings. Some research indicates that the accuracy
and performance of ordinary SVM can be improved by using the
Ensemble method [7]. The Ensemble method is a method that
combines the predictions generated by some classifier. One of
the Ensemble Methods of a classification is boosting algorithm.
which is more popular used compared to bagging algorithm [7].
So, in this research will compare the performance of ordinary
SVM with SVM using adaboost algorithm.
II. LITERATURE
As for some related research that discusses the predictions of
potential customers to subscribe deposits are as follows. Moro
(2011), et al. [2] proposed Cross Industry Standard Process for
Data Mining (CRISP-DM) method indicates that the support
vector machine gets better AUC and ALIFT results than the
Decision Tree and Naive Bayes methods. Feng, et al. [7]
proposed a Bayesian networks method and their application
shows that a single support vector machine method only
produces an average level of accuracy that is not much different
than other classification methods, and thus they conducted
ensemble method to improve the accuracy of the classification
method. Moro (2014), et al. [8] proposed a data-driven approach
to predict the success of bank telemarketing by comparing four
methods of data mining classification, i.e., Decision Tree,
Neural Networks, Support Vector Machine and Logistic
Regression. Their result showed that neural network is the best
*)
Corresponding Author: arvasit@gmail.com

Name Type
Age Numeric
Job Categorical : "Admin.", "Blue-Collar",
"Entrepreneur", "Housemaid",
"Management", "Retired", "Self-
Employed", "Services", "Student",
"Technician", "Unemployed", "Unknown"
Marital Categorical : "Divorced", "Married",
"Single", "Unknown";
Education Categorical : "Basic.4y", "Basic.6y",
"Basic.9y", "High.School", "Illiterate",
"Professional.Course",
"University.Degree", "Unknown"
Default Categorical: "No","Yes","Unknown"
Housing Categorical: "No","Yes","Unknown"
Loan Categorical: "No","Yes","Unknown"
Contact Categorical: "Cellular","Telephone"
Month Categorical: "Jan", "Feb", "Mar", ...,
"Nov", "Dec"
Day_Of_Week Categorical: "Mon", "Tue", "Wed", "Thu",
"Fri"
Duration Numeric
Campaign Numeric
Pdays Numeric
Previous Numeric
Poutcome Categorical : "Failure", "Nonexistent",
"Success"
Emp.Var.Rate Numeric
Cons.Price.Idx Numeric
Cons.Conf.Idx Numeric
Euribor3m Numeric
Nr.Employed Numeric
Y Binary: "Yes","No"
performance, and support vector machines and neural networks
are more flexible and have better learning skills among other
classification methods. Alhakbani and Rifaie [9] proposed a
handling class imbalance in direct marketing dataset using a
hybrid data. An algorithmic level solution using the Syntethic
Minority oversampling technique is implemented to balance the
dataset and then use grid search to optimize the c, gamma, and
kernel parameters. The models used the support vector machine
to build the HybridDA model. The results obtained indicate that
the technique is better than the techniques used by previous
research.
III. PRELIMINARIES
The datasets used in this paper is taken from the University
of California at Irvine (UCI) Machine Learning Repository. It is
a bank marketing data collected from March 2008 to November
2010. The number of attributes is 20 attributes with the addition
of 1 target class, and the total observation is 41,188 data [8].
Table I. Attributes Description
A. Support Vector Machine
Support Vector Machine (SVM) is one of the methods used
in the problem of pattern classification (pattern classification).
The basic idea of the SVM method is to maximize the
boundary-field separator (hyperplane) that separates the data
into two classes in a feature space. Support vector machine
method is suitable for use on linearly separable data [10].
To classifying the data cannot be separated linearly and it
should be transformed into feature space dimension such that
the data can be separated linearly from the feature space.
Feature space in practice usually has a higher dimension of
input space. Suppose there is a dataset whose data has two
attributes and two classes of positive and negative classes when
depicted in two-dimensional space of data cannot be separated
linearly. Therefore, the data is transformed to a higher
dimension so that it can be separated linearly. This method is
called kernel methods, which transform a linear SVM into non-
linear SVM
B. Adaboost
Boosting is a common and effective method for building an
accurate classifier by combining weak classifiers [7] The use of
boosting is preferred because it focuses on misclassified issues
and has a tendency to increase higher accuracy compared to
bagging. The commonly used boosting algorithm is the adaboost
algorithm
Adaboost trains the basic classifier sequentially (iteratively)
each iteration. The basic classification is trained by using data
training with weight coefficients that depend on the classifier
performance. On the previous iteration to give greater weight to
the wrongly classified data (misclassified). If the classifier has
been trained as much as desired, then the classifier is combined
to form a final decision on the model that shows the best
performance
The steps of the Adaboost algorithm are [11]
1. Input : Training data along with its label
, Component Learner, And the
number of iterations of T.
2. Initialization of training data weight
. (1)
3. For iteration
a. Use the Component Learner algorithm to train a
classification component, , On training weights.
b. Calculate the weight of classification error on
. (2)
The learning trust index is calculated as:
. (3)
c. Renew the sample training weight
. (4)
d. Testing model with testing data

Method Iteration Accuracy Sensitivity
Ordinary SVM - 91,67 % 83,80 %
Adaboost SVM
10 92,53 % 85,93 %
20 94,30 % 89,68 %
30 95,07 % 91,65 %
40 94,81 % 91,05 %
50 94,86 % 91,05 %
4. The last learning output
Combination of all classifications
Table III. Classification results of SVM Adaboost
C. Performance Evaluation
. (5)
The performance result of the proposed method is evaluated
based on the degree of accuracy and sensitivity. The
measurement evaluation is obtained from the confusion matrix
which consists of 4 parts of the classification results, i.e., True
Positive (TP), True Negative (TN), False Positive (FP) and False
Negative (FN). The confusion matrix is given in Table II.
Table II. Confusion Matrix/ Contingency Table
Actual Prediction True False
True TP FN
False FP TN
The accuracy level is defined as the degree of proximity
between the predicted data with the actual data or the ratio of the
amount of data that is correctly classified as shown by equation
(6).
Table III shows that the adaboost SVM method gives higher
accuracy rate since the algorithm updates the weights of
incorrect classified data to the specified iteration limit. The table
indicates that the increasing number of iterations is not
guaranteed the performance obtained will be better. In this
result, the iteration 30th
gives the best performance.
VI. CONCLUSION
In this paper proposed adaboost SVM method to improve
prediction results of direct marketing of bank dataset compare
to the ordinary SVM. The Adaboost algorithm successfully
Accuracy rate . (6) improves the performance of the SVM method. The results
The sensitivity level is defined as the ratio of data classified
positively [12]given by the equation (7).
obtained shows that the accuracy of the ordinary SVM only get
91.67% and sensitivity of 83.80%, whereas our proposed
adaboost SVM gives accuracy to 95,07% and sensitivity
th
Sensitivity . (7)
91,65% in the 30 iteration.
A. Dataset
IV. EXPERIMENTAL DESIGN
REFERENCES
[1] S. Abbas, “Deposit subscribe Prediction using Data Mining Techniques
based Real Marketing Dataset,” Int. J. Comput. Appl., vol. 110, no. 3, pp.
The total of 41,188 data will gets better accuracy results;
however, since the amount of data on the target deposit
subscribed customers is only 4,640 data which is less than
compared with customers who do not subscribed of 36,548 data.
This sitation will tend to predict the new data more incline to
deposit unsubscribe costumer, and therefore, in this experiment,
the amount of data is balanced to obtain the same probability
outcome. Hence, in this experiment use 9,280 total data with the
amount of comparison of training data is 70% and testing data is
30%.
B. Normalization data
For each attribute of type "categorical" will be changed to
"numerical". This is because data other than the "numerical"
type cannot be processed. The next type of target is changed to
1 for "yes" class and -1 for "No" class
C. Implementation
In this research will compare the performance results from a
single vector support method with support vector machine using
the adaboost algorithm.
V. RESULTS
The performance comparison result of performance of
ordinary SVM and adaboost SVM is given in Table III.
975–8887, 2015.
[2] S. Moro and R. M. S. Laureano, “Using Data Mining for Bank Direct
Marketing: An application of the CRISP-DM methodology,” Eur. Simul.
Model. Conf., no. Figure 1, pp. 117–121, 2011.
[3] C. Vajiramedhin and A. Suebsing, “Feature Selection with Data Balancing
for Prediction of Bank Telemarketing,” Appl. Math. Sci., vol. 8, no. 114,
pp. 5667–5672, 2014.
[4] H. A. Elsalamony, “Bank Direct Marketing Analysis of Data Mining
Techniques,” Int. J. Comput. Appl., vol. 85, no. 7, pp. 12–22, 2014.
[5] H. Elsalamony and A. Elsayad, “Bank Direct Marketing Based on Neural
Network,” Int. J. Eng. Adv. Technol., vol. 2, no. 6, pp. 392–400, 2013.
[6] L. Zhou, K. K. Lai, and L. Yu, “Least squares support vector machines
ensemble models for credit scoring,” Expert Syst. Appl., vol. 37, no. 1, pp.
127–133, 2010.
[7] G. Feng, J.-D. Zhang, and S. Shaoyi Liao, “A novel method for combining
Bayesian networks, theoretical analysis, and its applications,” Pattern
Recognit., vol. 47, no. 5, pp. 2057–2069, 2014.
[8] S. Moro, P. Cortez, and P. Rita, “A data-driven approach to predict the
success of bank telemarketing,” Decis. Support Syst., vol. 62, pp. 22–31,
2014.
[9] H. A. Alhakbani and M. M. Al-Rifaie, “Handling Class Imbalance in Direct
Marketing Dataset using A Hybrid Data and Algorithmic Level Solutions,”
SAI Comput. Conf. 2016, pp. 1–6, 2016.

[10]C. J. C. Burges, “A Tutorial on Support Vector Machines for Pattern
Recognition,” Data Min. Knowl. Discov., vol. 2, no. 2, pp. 121–167, 1998.
[11]Y. Freund, “A more robust boosting algorithm,” vol. arXiv:0905, 2009.
[12]G. Nisbet, R., Elder, J., Miner, Handbook of statistical analysis and data
mining applications. 2009.

direct marketing in banking using data mining

More Related Content

What's hot (19)

Similar to direct marketing in banking using data mining (20)

Recently uploaded (20)

direct marketing in banking using data mining