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