International Journal of Mechanical Engineering and Technology (IJMET)
Volume 10, Issue 01, January 2019, pp. 935–946, Article ID: IJMET_10_01_096
Available online at http://www.iaeme.com/ijmet/issues.asp?JType=IJMET&VType=10&IType=01
ISSN Print: 0976-6340 and ISSN Online: 0976-6359
© IAEME Publication
Scopus Indexed
NTRS: A NEW TRAVEL RECOMMENDATION
SYSTEM FRAMEWORK BY HYBRID DATA
MINING
T Uçar
Asst.Prof.Dr. Bahcesehir University, Software Engineering Department, Turkey
A Karahoca
Professor, Bahcesehir University, Software Engineering Department, Turkey
D Karahoca
Assistant Professor, Bahcesehir University, Health Management Department, Turkey
ABSTRACT
This study suggests a new travel recommendation system (NTRS) that was developed
to generate alternative travel destinations for customers. The proposed approach employs
hybrid data mining methods on NTRS by combining classification and clustering
algorithms. NTRS can be used for different travel data resources to find the best
prediction model to generate accurate recommendations. NTRS was tested by using real
travel data which contains flight and hotel bookings. Before applying data mining
algorithms, data set was cleansed, grouped and preprocessed. Then classification
techniques; ANFIS, RBFN and Naïve Bayes were combined with X-means and Fuzzy Cmeans clustering algorithms to find the best prediction model for proposing alternative
trips via NTRS. To identify the most suitable prediction model; recall, specificity,
precision, correctness, and RMSE scores were benchmarked and the best one was
dynamically selected. According to the testing scenario results, ANFIS and X-means
combination scored the finest RMSE and correctness values. Based on the proposed
approach’s algorithm, travel locations including trip durations and airline companies
were generated as recommendation output of the testing scenario. Generated
recommendation items can be used for providing suggestions for individuals or it can be
used by travel agencies for planning travel campaigns for target traveler groups. NTRS
proves that it can be executed for different data sets with hybrid data mining methods.
Key words: Recommender systems, data mining, travel planning and hybrid data mining
methods.
Cite this Article: T Uçar, A Karahoca and D Karahoca, Ntrs: a New Travel
Recommendation System Framework by Hybrid Data Mining, International Journal of
Mechanical Engineering and Technology, 10(01), 2019, pp.935–946
http://www.iaeme.com/IJMET/issues.asp?JType=IJMET&VType=10&Type=01
http://www.iaeme.com/IJMET/index.asp
935
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
1. INTRODUCTION
Information filtering systems are one of the popular research topics in the field of data mining.
Recommender systems are major applications for this area. They are used for filtering and
providing relevant information about a person’s search on a specific topic. To provide accurate
results, different data mining methods can be used [1-3].
Basically, a recommender system tries to generate a rating value of an item for a target user.
And according to rating values, system tries to propose an item or many items to target user. For
different users, every item gets associated with different rating scores. Rating scores can be
computed in various ways. In many recommender systems, target user’s profile and previous
behaviors are used for generating rating scores.
Recommender systems are being used in almost every search related area. Tourism domain
is one of these sectors. Most of the tourism-related recommender system applications involve
proposing travel schedules and location-based travel suggestions within a given set of user
defined constraints such as budget limits, time intervals, interests, desired locations or similar
necessities. After retrieving restrictions, a recommender system analyzes user input and proposes
a relevant output. To generate accurate predictions, many approaches are tested by different
researchers. In general, most of these approaches are based on acquiring a set of parameters which
can be used as constraints for the recommendation system [4], [5].
Hybrid data mining models are not used in travel recommendation engines. In this study,
hybridization means combining clustering and classification methods respectively to predict most
suitable locations for the travelers based on their travel behaviors. This approach was inspired
from a hybrid data mining solution that was used for predicting GSM churners [6]. Also, in
literature, hybrid data mining models are used to improve the effectiveness and efficiency of
predictive models in different areas such as customer retention [7], credit scoring [8], churn
management [9], heart diseases [10].
This study proposes a recommender system framework which can accurately classify
travelers. Based on classification output, possible travel destinations, trip durations and
transportation companies are recommended. This recommendation output can be used to help
customers who are planning trips, or it can be used by travel agencies to plan travel campaigns
for similar users.
The remainder of this document is organized as follows: Section 2 includes a brief review of
recent studies about recommender systems, Section 3 describes materials and methods used in
the proposed approach, Section 4 presents output of the test scenario results and Section 5
contains conclusion and possible future plans of this study.
2. BACKGROUND
Travel recommender studies can be categorized into three main groups which are proposing
recommendations for preparing travel schedules, proposing tour plans based on user preferences
and proposing recommendations by extracting information from social media-based sources.
Mainly, a travel schedule planner tries to generate a time-table based trip recommendation by
considering user-provided set of constraints. In the literature, many studies were conducted using
different approaches. A mobile software solution is developed in [11]. An ontology-based web
recommender was developed in [12]. A trip recommender which enhances hotel services was
proposed in [13].
Tour planners serve a similar purpose as tour schedule planners. But unlike schedule planners,
they do not employ a time-table based recommender. Lucas et al. [14] proposed a hybrid
recommendation system which contains both content-based and collaborative filtering methods
using association-based classification approach. Yang and Hwang [15] used a location-based
http://www.iaeme.com/IJMET/index.asp
936
editor@iaeme.com
Ntrs: a New Travel Recommendation System Framework by Hybrid Data Mining
collaborative filtering method for the implementation of a tour planner. Vukovic and Jevtic [16]
studied on a location predictor for mobile users. Varfolomeyev et al. [17] proposed a
recommender system for historical tourism. Colomo-Palacios at al. [18] studied on a contextaware mobile recommendation system for loyalty in tourism.
Unlike the previous two recommender groups, a social media-based recommender takes
advantage of social networks to generate recommendations. Umanets et al. [19] presented a
mobile and a web-based recommender system for tourists which can integrate with social
networks. Han and Lee [20] implemented a recommendation system which analyzes geo-tagged
social media to recommend landmarks for customized travel planning. García-Palomares et al.
[21] presented a method for identifying tourist attractions in cities using patterns retrieved from
geo-tagged photos. Sun et al. [22] developed a recommender system which analyzes photos from
Flickr to discover popular tourist locations and possible travel destinations between them.
Based on the reviewed literature above, the proposed recommendation approach in this study
differs from the others by generating travel destination recommendations not by considering a
fixed method or method combination, but by dynamically comparing and combining clustering
and classification algorithms using the provided data.
The next section presents the applied data mining methods along with data gathering and
processing steps of the sample data set which was used to test the proposed approach.
3. MATERIAL AND METHODS
In this study, flight and hotel booking data was used for testing the proposed recommendation
model. Detailed information about this data set and applied algorithms are stated below.
3.1. Data Gathering and Processing
Initial data set of customer flights and hotel bookings was obtained from an existing travel
platform which is integrated to different booking service provider companies. Each company
returns booking data in XML and the travel platform used a relational database to keep these
values.
Data gathering process involved two steps: Step 1) Running T-SQL blocks to retrieve XML
formatted data from database tables. Step 2) Running XPath expressions on the retrieved
attributes to extract actual values from XML formatted data.
After running the described steps above, a total of 26886 flight records and 4367 hotel
bookings were collected. Retrieved fight and hotel booking records were joined and 317 matching
entries were obtained. These matching entries contained the booking records of customers where
a flight and a corresponding hotel reservation was made by the same person. All of the identity
columns were removed from the obtained data set. At the end of this data processing, 14 attributes
were gathered as listed in Table 1.
http://www.iaeme.com/IJMET/index.asp
937
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
Table 1 Initial data set attributes
Attribute
Gender
Departure date
Description
Passenger’s gender.
Starting date of travel.
Departure location
Arrival location
Location which the passenger is leaving form.
Location which the passenger is arriving to.
Departure airline
Departure flight class
Returning date
Airline company for departure flight.
Ticket class for departure flight.
Ending date of travel.
Returning location (from)
Returning location (to)
Returning airline
Location which the passenger is returning from.
Location which the passenger is returning to.
Airline company for returning flight.
Returning flight class
Flight cost
Ticket class for returning flight.
Flight’s cost.
Days in hotel
Hotel cost
Number of days stayed in hotel.
Hotel’s cost.
Attributes in Table 1 are the only attributes revealed by the travel platform to use in this study.
“Departure location” attribute was removed from the initial data set since “departure location”
and “returning location (to)” attributes were containing the same set of values. “Arrival location”
and “returning location (from/to)” values were discretized according to regions. Table 2 lists
regions by their numeric codes.
Table 2 Region codes
Code
1
Description
Northern Europe
Code
12
Description
Central Asia
2
3
4
Southern Europe
Eastern Europe
Western Europe
13
14
15
Africa
America
Australia
5
6
Central Europe
Balkans
16
17
(Turkey) Marmara Region
(Turkey) Black Sea Region
7
8
9
Middle East
Northern Asia
Southern Asia
18
19
20
(Turkey) Central Anatolia Region
(Turkey) Southeastern Anatolia Region
(Turkey) Aegean Region
10
11
Eastern Asia
Western Asia
21
22
(Turkey) Eastern Anatolia Region
(Turkey) Mediterranean Region
Based on travelers’ flight and hotel expenses, cost attributes were discretized into six groups.
Table 3 and Table 4 lists cost attribute groups.
Table 3 Flight cost groups
Code
1
Description
< 200
2
3
201 – 400
401 – 700
4
5
6
701 – 1400
1401 – 3000
4000 +
http://www.iaeme.com/IJMET/index.asp
938
editor@iaeme.com
Ntrs: a New Travel Recommendation System Framework by Hybrid Data Mining
Table 4 Hotel cost groups
Code
1
2
3
4
5
6
Description
< 350
351 – 700
701 – 1000
1001 – 1500
1501 – 2500
2500 +
Travel season and travel duration values were derived using “departure date” and “returning
date” attributes. “Days in hotel” attribute was removed from data set because travel duration and
“days in hotel” values were pointing to same set of values. Ticket class attributes were also
removed from the data set since 97% of records were sharing the same ticket class type. After
deriving these two new attributes, removing redundant attributes and applying discretization on
initial data set, 10 attributes were obtained for data processing. Table 5 lists the final state of these
attributes.
Table 5 Final data set attributes
Attribute
Gender
Travel duration
Season
Arrival location
Departure airline
Returning location (from)
Returning location (to)
Returning airline
Flight cost
Hotel cost
Description
Passenger’s gender.
Duration of travel in days.
Season of travel.
Location which the passenger is arriving to.
Airline company for departure flight.
Location which the passenger is returning from.
Location which the passenger is returning to.
Airline company for returning flight.
Flight’s cost.
Hotel’s cost.
Final data set was used to run data mining algorithms. 66% of data was used for training and
the remaining 34% was used for testing the prediction models.
3.2. Methods
K-means, Fuzzy C-means, Adaptive Neuro Fuzzy Inference System, Radial Basis Function
Networks and Naïve Bayes methods are briefly explained in the following sub-sections.
3.2.1. X-Means (XM) Clustering
K-means clustering algorithm is a simple but popular method for finding clusters in a given data
set. However, there are some important shortcomings for this method such as the necessity of
providing the number of clusters and random located initial cluster centers. XM can be
summarized as an improved version of the K-means clustering algorithm with self-estimation of
the number of clusters for a given data set [23].
3.2.2. Fuzzy C-means (FCM) Clustering
FCM is a soft clustering algorithm where each data point in a data set has a degree of belonging
to clusters. For any point x, there is a set of coefficients which gives the degree of being a member
for a given cluster. As an outcome of this membership degrees, points at the edges of clusters can
http://www.iaeme.com/IJMET/index.asp
939
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
be shared by other clusters. The degree of belonging is related inversely to the distance from x to
the cluster center [24], [25].
3.2.3. Adaptive Neuro Fuzzy Inference System (ANFIS)
ANFIS employs both neural networks and fuzzy systems for proposing a neural-fuzzy system. A
fuzzy-logic system basically maps the input space to the output space in a non-linear way. To
perform this type of mapping, numerical inputs of the system are converted to fuzzy domain using
fuzzy sets and fuzzifiers. After this step, the obtained fuzzy domain gets applied with fuzzy rules
and fuzzy inference engine. This process produces a result where defuzzifiers are used for
converting it back to arithmetical domain. Gaussian functions are used for fuzzy sets and linear
functions are used for rule outputs on ANFIS method. Network parameters of the system are
obtained by computing coefficients of the output linear functions, mean of the membership
functions and standard deviation. The last node of the system which is the rightmost node of a
network contains the calculation of summation of each output [26].
3.2.4. Radial Basis Function Networks (RBFN)
RBFN is a popular feedforward neural network model. It contains three layers including the input
layer. Each point in input space is represented by a hidden unit. Output (activation) of a hidden
unit is based on the distance between the hidden unit’s point and the instance. The activation for
a given hidden unit will be stronger for closer points. A nonlinear transformation function is
required for converting a distance into a similarity measure. A Gaussian activation function is the
mostly used transformation function for this requirement. For each hidden unit, the bell-shaped
width of Gaussian activation function can be different. The hidden units are called as Radial Basis
Functions. Radial basis function network’s output layer works as similar as multilayer perceptron.
Output of hidden units is received as a linear combination and it runs them through the sigmoid
function [27].
3.2.5. Naïve Bayes (NB)
Naïve Bayes is a simple but highly scalable probabilistic classifier which is built on Bayes’
theorem. It assumes that the value of an attribute is independent of the value of any other attribute.
Let x be a data vector, h be a hypothesis for x to be member of class c. P(h│x) is the probability
of x to be a member of c which is called as posterior probability. P (h) is called as the prior
probability which is independent of x. To classify an instance vector x = (x1, x2, xn) having n
attributes, Naïve Bayes classifier predicts that x belongs to the class which has the highest
posterior probability conditioned on x among m classes. Maximum posteriori hypothesis is the
class ci for which P (c_i│x) is maximized. Probabilities of a given data being a member of a class
can be easily estimated using training data set [28].
3.3. Comparing Data Mining Methods
Benchmarking and comparing different data mining techniques can be done by computing
confusion matrix for each method. The simplest confusion matrix can be constructed for binary
classification problems where output is mapped to two clusters. For such problems the
constructed confusion matrix will be a two-dimensional square matrix. For non-binary
classification problems, it will be an n-dimensional square matrix. In an n-dimensional confusion
matrix, row indices represent actual values whereas column indices represent predicted values
for a classification problem. Figure 1 shows the structure of a binary confusion matrix.
http://www.iaeme.com/IJMET/index.asp
940
editor@iaeme.com
Ntrs: a New Travel Recommendation System Framework by Hybrid Data Mining
Figure 1. Sample binary confusion matrix
To construct a confusion matrix, true positive (TP), false negative (FN), false positive (FP) and
true negative (TN) values are required. TP is the number of positive examples correctly predicted
by the classification model. FN is the number of positive examples wrongly predicted as negative
by the classification model. FP is the number of negative examples wrongly predicted as positive
by the classification model and TN is the number of negative examples correctly predicted by the
classification model. True positive rate (TPR) which is also called as recall is the fraction of
positive examples predicted correctly by the classification model.
(
)=
(
)
+
(1)
True negative rate (TNR) which is also called as specificity is the fraction of negative examples
predicted correctly by the classification model.
=
(
)
+
(2)
Precision is the ratio of true positive instances by the total number of true positive and false
positive instances.
=
(
+
)
(3)
Correctness is the percentage of correctly classified instances by the classification model. Root
mean squared error (RMSE) is used for measuring the differences between actual and predicted
instances. Equation 4 shows the RMSE formulation where n is the number of total instances, p is
the predicted values and r is the actual values.
=
∑
(
− )
(4)
3.4. The Proposed Approach
Existing recommender systems mainly employ data mining solutions for generating
recommendations as stated in background section. Since data used in these studies are different
from each other, researchers tried to find the best methods or algorithms based on their data set.
But when a different data set with different data features is used, these approaches cannot assure
providing accurate recommendations. This fact emerges the need of an expert system which can
dynamically adapt itself to different data sets.
http://www.iaeme.com/IJMET/index.asp
941
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
Instead of using a fixed data mining solution, the proposed approach in this study builds
various prediction models using different classification algorithms on differently clustered
versions of the provided data set. Clustered data set versions are obtained by applying different
clustering algorithms with different cluster sizes. Generated classification models are compared
by prediction correctness scores. The classification and clustering method pair which yields the
highest correctness score is picked for generating recommendations. Algorithm 1 shows this
process.
The first step of the proposed approach is the clustering phase. In this step, five different
versions of the same data set are created by clustering into four to eight groups. When the first
part ends, the second step is finding the best classification algorithm / clustered data pair which
will be used for recommendations. A fuzzy inference system, a neural network model and a
probabilistic classifier is used on the generated versions of the data set to generate different
prediction models. The model with the highest correctness score is designated as the primary
model of the recommender system. The third step is predicting the cluster of a given input
(customer) by using the designated model which was picked in the previous step. After predicting
the input’s cluster, the fourth step makes recommendations based on similar users from the same
cluster. To perform this task, system creates a list of possible destination locations by finding
previously preferred destinations of instances which are from the same cluster. Then it computes
frequencies of each location and picks three locations from this list by finding top three frequency
values. For each selected location, system finds the most preferred airline company and average
travel duration values. Figure 2 illustrates each of these steps and Algorithm 2 shows the process
of generating recommendation items.
Algorithm 1. Finding the best prediction model
01:
initialize highest correctness model container with the first model in prediction models
02:
foreach model in prediction models do
03:
if model’s correctness exceeds highest correctness model’s correctness then
04:
update highest correctness model
05:
end
06:
end
Algorithm 2. Generating recommendation items
01:
get initial input from user
02:
classify user’s cluster based on retrieved input using the best prediction model
03:
foreach user in classified user’s cluster do
04:
add user's trip duration, trip destination and airline values to raw recommendation list
05:
end
06:
foreach destination in raw recommendation list do
07:
construct top three preferred destinations list
08:
end
09:
foreach destination in top three preferred destinations list do
10:
find most preferred airline company for destination
11:
find average trip duration for destination
12:
bind destination with most preferred airline and average trip duration properties
13:
end
14:
set top three destinations and bound properties as recommendation output
http://www.iaeme.com/IJMET/index.asp
942
editor@iaeme.com
Ntrs: a New Travel Recommendation System Framework by Hybrid Data Mining
Figure 2. The proposed approach
Figure 3 shows an overview of the connections among then entire framework modules.
Figure 3. Recommender system framework
4. RESULTS
The proposed recommendation model was tested using the processed traveler data set which was
clustered into four to eight clusters using XM and FCM algorithms. Then, ANFIS, RBFN and
NB classification algorithms were applied to find the prediction model with the highest
correctness value.
Table 6 shows benchmarking results for each differently clustered data set and method
combination which was obtained as an outcome of this test scenario.
http://www.iaeme.com/IJMET/index.asp
943
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
Table 6 Benchmarking Methods for testing
Classifier
Clusterer
ANFIS
ANFIS
ANFIS
ANFIS
ANFIS
RBFN
NB
NB
RBFN
RBFN
NB
RBFN
NB
NB
RBFN
NB
RBFN
NB
NB
RBFN
NB
RBFN
NB
RBFN
ANFIS
RBFN
ANFIS
ANFIS
ANFIS
ANFIS
XM
XM
XM
XM
XM
FCM
FCM
XM
FCM
FCM
XM
XM
XM
FCM
XM
FCM
FCM
FCM
FCM
XM
XM
FCM
XM
XM
FCM
XM
FCM
FCM
FCM
FCM
Cluster
Size
8
5
6
7
4
5
8
8
4
8
4
8
7
7
6
4
6
5
6
7
5
7
6
4
6
5
4
5
7
8
Recall
Specificity
Precision
Correctness
RMSE
0.86
0.89
0.86
0.81
0.93
0.96
0.92
0.94
0.96
0.93
0.95
0.93
0.92
0.93
0.93
0.94
0.93
0.93
0.93
0.91
0.92
0.89
0.85
0.93
0.54
0.89
0.71
0.63
0.58
0.42
0.98
0.98
0.97
0.98
0.97
0.99
0.99
0.99
0.99
0.99
0.98
0.98
0.98
0.99
0.97
0.98
0.98
0.98
0.98
0.98
0.98
0.98
0.95
0.98
0.91
0.96
0.89
0.91
0.92
0.92
0.85
0.94
0.90
0.83
0.95
0.97
0.92
0.94
0.97
0.94
0.95
0.93
0.92
0.93
0.93
0.95
0.93
0.93
0.93
0.92
0.92
0.91
0.86
0.94
0.53
0.90
0.69
0.65
0.51
0.41
97.22
97.04
95.99
96.30
96.76
96.30
91.67
93.52
96.30
92.59
95.37
92.59
91.67
92.59
92.59
94.44
92.59
92.59
92.59
90.74
91.67
88.89
85.19
92.59
85.19
88.89
83.33
85.19
85.98
86.57
0.07
0.07
0.07
0.08
0.08
0.11
0.11
0.11
0.12
0.13
0.13
0.14
0.14
0.14
0.15
0.15
0.15
0.15
0.16
0.16
0.18
0.18
0.18
0.19
0.20
0.21
0.21
0.23
0.23
0.28
Based on these values, ANFIS has the highest RMSE and correctness scores when applied on
data set which is clustered with X-means into eight clusters. RBFN generates the highest recall,
specificity and precision values when applied on data set which is clustered with Fuzzy C-means
into five clusters. Naïve Bayes presents its highest specificity value when applied on data set
which is clustered with Fuzzy C-means or X-means into eight clusters.
If we compare the correctness and RMSE values between these three methods, we can state
that top five scores are obtained by ANFIS and X-means combination where the highest model
correctness is 97.22 with an RMSE value of 0.07.
The benchmark values in Table 6 are specifically generated for the test scenario data set which
means that for a different data set, system will behave in a different way to achieve the highest
correctness score for generating recommendations.
http://www.iaeme.com/IJMET/index.asp
944
editor@iaeme.com
Ntrs: a New Travel Recommendation System Framework by Hybrid Data Mining
5. CONCLUSION
Predicting possible travel destinations for travelers can be very advantageous especially for travel
agencies. When possible travel destinations along with travel durations are combined, such
information can be used for defining package tour plans which can be offered by travel companies
to a user or similar users. And proposing possible airline services for the suggested travel plan
can even make the proposed program more beneficial.
From this viewpoint, unlike previously implemented systems, the proposed recommender
system framework employs a novel prediction engine for discovering similar travelers. It
generates possible travel destinations based on users. The defined approach can adapt itself to
different data sets since it has a hybrid and dynamic design.
No hotel suggestions are available in this study. But, proposing possible hotels for the
predicted cluster’s hotel price range by adding user specified constraints (like proximity to city
center, being near to shore, etc.) can be a future extension for the proposed framework.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
Sharma, P. K. & Sinha, A. K. (2018). Development of an Intelligent Manet System Model for
Detection and Prevention of Annomaly Using Anfis, International Journal of Mechanical
Engineering and Technology 9(3), 301-312.
Lü, L., Medo, M., Yeung, C. H., Zhang, Y. C., Zhang, Z. K., & Zhou, T. (2012).
Recommender systems. Physics reports, 519(1), 1-49.
Liao, S. H., Wen, C. H., Hsian, P. Y., Li, C. W., & Hsu, C. W. (2014). Mining Customer
Knowledge for a Recommendation System in Convenience Stores. International Journal of
Data Warehousing and Mining (IJDWM), 10(2), 55-86.
Borràs, J., Moreno, A., & Valls, A. (2014). Intelligent tourism recommender systems: A
survey. Expert Systems with Applications, 41(16), 7370-7389.
Sun, H., Huang, J., She, X., Yang, Z., Liu, J., Zou, J., ... & Wang, D. (2015). TripRec: An
Efficient Approach for Trip Planning with Time Constraints. International Journal of Data
Warehousing and Mining (IJDWM), 11(1), 45-65.
Karahoca, A., & Karahoca, D. (2011). GSM churn management by using fuzzy c-means
clustering and adaptive neuro fuzzy inference system. Expert Systems with Applications,
38(3), 1814-1822.
Chu, B. H., Tsai, M. S., & Ho, C. S. (2007). Toward a hybrid data mining model for customer
retention. Knowledge-Based Systems, 20(8), 703-718.
Koutanaei, F. N., Sajedi, H., & Khanbabaei, M. (2015). A hybrid data mining model of feature
selection algorithms and ensemble learning classifiers for credit scoring. Journal of Retailing
and Consumer Services, 27, 11-23.
Hudaib, A., Dannoun, R., Harfoushi, O., Obiedat, R., & Faris, H. (2015). Hybrid data mining
models for predicting customer churn. International Journal of Communications, Network
and System Sciences, 8(05), 91.
Verma, L., Srivastava, S., & Negi, P. C. (2016). A hybrid data mining model to predict
coronary artery disease cases using non-invasive clinical data. Journal of medical systems,
40(7), 178.
Castillo, L., Armengol, E., Onaindía, E., Sebastiá, L., González-Boticario, J., Rodríguez, A.,
... & Borrajo, D. (2008). SAMAP: An user-oriented adaptive system for planning tourist
visits. Expert Systems with Applications, 34(2), 1318-1332.
Moreno, A., Valls, A., Isern, D., Marin, L., & Borràs, J. (2013). Sigtur/e-destination:
ontology-based personalized recommendation of tourism and leisure activities. Engineering
Applications of Artificial Intelligence, 26(1), 633-651.
http://www.iaeme.com/IJMET/index.asp
945
editor@iaeme.com
T Uçar, A Karahoca and D Karahoca
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
Silamai, N., Khamchuen, N., & Phithakkitnukoon, S. (2017, September). TripRec: trip plan
recommendation system that enhances hotel services. In Proceedings of the 2017 ACM
International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of
the 2017 ACM International Symposium on Wearable Computers (pp. 412-420). ACM.
Lucas, J. P., Luz, N., Moreno, M. N., Anacleto, R., Figueiredo, A. A., & Martins, C. (2013).
A hybrid recommendation approach for a tourism system. Expert Systems with Applications,
40(9), 3532-3550.
Yang, W. S., & Hwang, S. Y. (2013). iTravel: A recommender system in mobile peer-to-peer
environment. Journal of Systems and Software, 86(1), 12-20.
Vukovic, M., & Jevtic, D. (2015). Agent-based Movement Analysis and Location Prediction
in Cellular Networks. Procedia Computer Science, 60, 517-526.
Varfolomeyev, A., Korzun, D., Ivanovs, A., Soms, H., & Petrina, O. (2015). Smart space
based recommendation service for historical tourism. Procedia Computer Science, 77, 85-91.
Colomo-Palacios, R., García-Peñalvo, F. J., Stantchev, V., & Misra, S. (2017). Towards a
social and context-aware mobile recommendation system for tourism. Pervasive and Mobile
Computing, 38, 505-515.
Umanets, A., Ferreira, A., & Leite, N. (2014). GuideMe–A tourist guide with a recommender
system and social interaction. Procedia Technology, 17, 407-414.
Han, J., & Lee, H. (2015). Adaptive landmark recommendations for travel planning:
personalizing and clustering landmarks using geo-tagged social media. Pervasive and Mobile
Computing, 18, 4-17.
García-Palomares, J. C., Gutiérrez, J., & Mínguez, C. (2015). Identification of tourist hot spots
based on social networks: A comparative analysis of European metropolises using photosharing services and GIS. Applied Geography, 63, 408-417.
Sun, Y., Fan, H., Bakillah, M., & Zipf, A. (2015). Road-based travel recommendation using
geo-tagged images. Computers, Environment and Urban Systems, 53, 110-122.
Pelleg, D., & Moore, A. W. (2000, June). X-means: Extending k-means with efficient
estimation of the number of clusters. In Icml (Vol. 1, pp. 727-734).
Bezdek, J. C., Coray, C., Gunderson, R., & Watson, J. (1981). Detection and characterization
of cluster substructure i. linear structure: Fuzzy c-lines. SIAM Journal on Applied
Mathematics, 40(2), 339-357.
Dunn, J. C. (1973). A fuzzy relative of the ISODATA process and its use in detecting compact
well-separated clusters.
Jang, J. S. (1993). ANFIS: adaptive-network-based fuzzy inference system. IEEE
transactions on systems, man, and cybernetics, 23(3), 665-685.
Witten, I. H., Frank, E., Hall, M. A., & Pal, C. J. (2016). Data Mining: Practical machine
learning tools and techniques. Morgan Kaufmann.
Han, J., Pei, J., & Kamber, M. (2011). Data mining: concepts and techniques. Elsevier.
http://www.iaeme.com/IJMET/index.asp
946
editor@iaeme.com