LDcFR: A new model to determine value of airline passengers

Abstract

The increasing willingness to travel as well as the increasing number of airlines create a strong competitive market in the air transport industry. In such circumstances, paying attention to the market needs is the prerequisite of the success and profitability of the company. Hence, nowadays, airlines endeavor to provide their customers with appropriate services and incentives to maintain the existing customers and attract new ones. The importance of providing appropriate incentives, on the one hand, and lack of resources, on the other, have led to airlines’ incapacity to provide good services for all of their customers. A logical solution in such a situation is to determine strategies to provide services for customers based on their value. The value model developed in this study is based on a combination of four variables, namely, relationship Length, Frequency, Distance and Recency, which is named LDcFR model. Using these variables along with the imperialist competitive algorithm clustering technique and Markov chain will determine the value of each passenger. The results of the study indicate that the LDcFR model is useful for determining the passenger value.

Keywords

Passenger value LDcFR imperialist competitive algorithm Markov chain customer relationship management

Introduction

Nowadays, the passengers are the most important asset of an airline. No bright future can be thought for an airline without satisfied passengers who will remain loyal and develop their relationship. In today’s complex and competitive environment in transportation industry, airlines tend to create programs to increase the level of satisfaction, maintain the current customers and attract new ones. Customer relationship management (CRM) has been developed for this purpose. The main goal of CRM is to create a long and profitable relationship with customers. This issue has been taken into consideration by many academic and practitioner researchers, and many industries have invested in this domain (Bohling et al., 2006; Sharma and Iyer, 2007). CRM is composed of a set of processes and enabling systems which support a business strategy to build long-term, profitable relationships with specific customers (Ling and Yen, 2001; Ngai et al., 2009). This process and system can lead to an increase in customer’s satisfaction and their loyalty, which in turn can result in retaining the existing customers and attracting new ones.

Implementation CRM is more important, especially, in service industries. This is because of the fact that in these industries, customer retention is far more important than customer attraction. Due to increasing the number of the travel service providers, customer satisfaction measurement is becoming an important strategy for many companies to achieve competitive advantages.

CRM suggests organizations to constantly monitor market needs to retain the existing customers and attract new ones. Furthermore, it offers new products or services based on the market needs. By providing new products or services based on customer needs, customers will be satisfied, and this satisfaction will make them loyal customers. It is noteworthy that the organizations cannot meet the needs of all customers. To solve this problem, the customers’ clustering according to their value is employed. In this case, the organization calculates customer’s value by using their features and records, and then clusters them. The best customers of any organization are the most valuable ones whom the organization endeavors to preserve. The organization identifies the most valuable class of its customers along with their needs and determines its future strategies and goals based on the respective needs. One of the most widely used methods for determining the customer value is using the Recency, Frequency, Monetary (RFM) model.

The rest of this paper is arranged as follows: Section “Literature review” provides an overview of the related literature, Section “Methodology” presents the proposed methodology, Section “Case study” states the experimental results of the study conducted in a real environment and the last section presents the conclusions.

Literature review

In this section, the CRM-related literature including RFM and the clustering technique will be summarized. Furthermore, imperialist competitive algorithm (ICA) and its application in customer clustering will be briefly described.

RFM

The RFM model, first proposed by Hughes in 1994, is one of the most renowned models which is widely applicable in customer clustering, CRM, customer value analysis and measuring customer lifetime value (Chang et al., 2010; Cheng and Chen, 2009; Hughes, 2006; Wei et al., 2012). The RFM model’s measurement system is on the basis of three variables that are shown in Table 1 (Akhondzadeh-Noughabi et al., 2013; Hu et al., 2013; Tabaei and Fathian, 2011).

Table 1.

RFM model.

R	Recency	Time elapsed since the customers’ last purchase to the present
F	Frequency	Number of purchases in a particular period.
M	Monetary	Total money spent by the customer in a particular period.

According to the above definition, the lower the value of “Recency” and the higher the values of “Frequency” and “Monetary” are, the more valuable the customer would be.

The RFM model measure is not persistently fixed and may be changed based on the specific characteristics of every industry and may become customized. For instance, in order to determine the customers’ values in the telecommunications industry, Cheng and Sun (2012) removed R from the RFM model and expanded the TFM (Time, Frequency, Monetary) Model (Huang et al., 2012). In another study, the development of RFM model has been used to classify the patients in a dental clinic. In this paper, the variable M is removed since monetary value (i.e. the cost of hospital service) to all patients is the same and instead the LRF model is presented. Variable L, expressing length of customer relationship, is the time of first visit to the last visit.

Consequently, nowadays, this model is used in various industries such as telecommunications, medicine, textile, health & cosmetics, financial institutions, etc. to calculate customer value (Chen et al., 2012; Cheng and Sun, 2012; Khajvand and Tarokh, 2011; Li et al., 2011; Wei et al., 2012; Yeh et al., 2009).

In transportation industry, the RFM model is also used to determine the value of passengers. Chiang (2012) developed FPDN model to determine the value of passengers in air transport industry. The measures of FPDN model include: Frequency, Price Discount, Destination and No-Show. Moreover, Chiang (2014) offered FMCN model to determine the value of airline passengers. The measures of FMCN model include: Frequency, Monetary, Cancelations and Number of family members. Furthermore, in some articles like Wong and Chung (2007) and Abbasi et al. (2010), simple RFM model and clustering algorithm have been utilized to determine passenger’s value.

Clustering

Clustering is one of the data mining techniques (Ngai et al., 2009) used for grouping data with homogeneous characteristics (Niknam et al., 2011). The goal of clustering technique is to incorporate the data with the most similarity into the same group (Akhondzadeh-Noughabi et al., 2013; Berry and Linoff, 2004; Niknam et al., 2011). Generally, there are two types of methods, namely, unsupervised and supervised, for clustering items and data. In the supervised method, the number of needed clusters is determined at the beginning of the procedure, whereas in the unsupervised method, no specific information is provided for the algorithm at the beginning of the phase. One of the most well-known supervised methods is the K-means, and one of the most renowned unsupervised methods is the Self Organizing Map (SOM). The K-means method is widely used for data clustering because of its simplicity, power, and speed (Chang et al., 2010; Niknam et al., 2011). In this method, the Euclidean is usually used to determine the distance.

Despite its simplicity and power of the clustering, K-means algorithm suffers from certain problems among which the following instances can be enumerated (Ho et al., 2012; Niknam and Amiri, 2010; Niknam et al., 2011; Zhongding et al., 2009)

Its dependency on the choice of initial centers.

Being influenced by the outliers.

Because of these problems, nowadays, evolutionary algorithms are used along with K-means. Due to the innovatory nature of the revolutionary algorithms, they can eliminate the above mentioned problems. A new evolutionary algorithm with high accuracy and speed is ICA that has been widely used in clustering processes.

Imperialist competitive algorithm

ICA is a kind of evolutionary algorithms that was introduced by Atashpaz-Gargari and Lucas (2007). ICA is inspired from the socio-political process to solve optimization problems in the real world (Gargari et al., 2008; Kaveh and Talatahari, 2010; Niknam et al., 2011; Rajabioun et al., 2008; Roshanaei et al., 2008). By emulating the countries’ social, economic and political process of evolution and by mathematically modeling of this process, this algorithm presents regular operators in the form of algorithm that can help to solve the complex problems of optimization.

Like other evolutionary algorithms, ICA algorithm also begins with forming a set of initial potential answers. In ICA, each of the initial answers is called a “country.” ICA aims to improve the countries and find the optimal answers. To achieve this goal, a specific process will be carried out within the algorithm. The main functions of this algorithm are assimilation policy, imperialist competition and revolution.

Niknam et al. (2011) combined ICA and K-means algorithm and developed a new approach (K-MICA) for clustering in which the problems of K-means algorithm have been resolved.

By comparing the results obtained from the developed algorithm with those of the other available algorithms, Niknam et al. (2011) showed that the K-MICA algorithm is faster and more accurate than the other algorithms. Hence, in this paper, the same method will be used in customers clustering.

Markov chain

A Markov chain is a probabilistic process that used to represent correlations between successive observations of a random variable. This sequence analysis process is a form of time-series modeling and was introduced at the beginning of the 20th century by Andrej Markov. Markov chain process has the memoryless property. Memoryless property in Markov chain means that the predicted state of the system only depends on the current state and earlier does not depend on the states. Equation (1) shows Markov chain Forgot property

P (X_{n + 1} | X_{n}, X_{n - 1}, \dots, X_{0}) = P (X_{n + 1} | X_{n})

(1)

One of the main applications of Markov chain is the probability of transition from one stage to the next step in the process. This chain, with review the past state of process, estimates the future possible states.

Methodology

In this section, a new methodology is presented based on RFM attributes and ICA to achieve the effective management of passengers in the airline industry. This framework is composed of three parts. In the first step, the RFM model is developed based on the characteristics of the transport industry, and in the second step, the clustering process will be carried out. In the last step, the Markov chain will be used in determining the passengers’ prospective value.

RFM model development

Special features of the transport industry and intense competition in the market have caused the airlines to use different strategies such as providing dynamic pricing, offering different services in the form of different classes with different charges and the kind, in order to attract and maintain the passengers. According to these strategies, the basic RFM model has lost its performance in determining the value of passengers. Hence, in this study, LDcFR model has been developed and replaced. The model variables of the present study are shown in Table 2.

Table 2.

LDcFR model variables.

Length (L)	Time elapsed since the first purchase to the last one.
Distance (Dc)	The total distance traveled by the customer (passenger) in a particular period.
Frequency (F)	Number of travels by a customer in a particular period.
Recency (R)	Time elapsed since the customers’ last purchase to the present.

As the definitions of the LDcFR model indication (Table 2), the monetary variable has been removed from the set of the RFM variables. Instead, Dc and L variables are replaced. The reason for this omission is the dynamicity and instability in the price of the ticket sales. Dynamicity and instability in price of airline ticket sales may be appeared in two forms:

Price dispersion in similar flights: Similar flights those that have the same origin and destination, but they can be done in various dates. In such conditions, the ticket price is subject to different factors such as “fuel price,” “degree of demand,” “seasonal conditions,” “market conditions,” etc. (Bisignani, 2010; Fedorco and Hospodka, 2013).

Price dispersion in the same flight: The “A flight price dispersion” refers to the non-uniformity of ticket prices for passengers during a specific flight. This instability of price in a flight can occurs due to different factors such as “difference in value of the passengers,” “capacity constraints,” “time of ticket purchasing,” etc. (Alderighi et al., 2011; KNEWS, 2013; Lindenmeier and Tscheulin, 2008).

There are also some other strategies for determining the ticket price which are used by different airline companies with respect to their particular circumstances. For instance, Samoa Air Airline Company determines its ticket prices according to the weight of the passengers and their luggage.

In such conditions, not only the prices vary from one airline to another but also the prices are different within an airline in different situations. Considering such conditions, it seems that the monetary variable cannot be an appropriate criterion for determining the value of passengers.

An important factor that can have a significant effect on the value of passengers is distance criterion. This criterion represents the distance traveled by passengers using the airline. Higher values of “Distance” indicate that the passenger is more valuable. What is important is the impact of flight’s class in calculating distance criteria. Class of travel is a term that refers to the amount of services provided and the passenger’s comfort during the trip. Class of travels are different from one another based on different factors such as the space of seats, number of flight attendants, flight attendants’ ability and commitment, the hospitality, entertainment services, etc. The more the level and quality of service offered and the more convenience provided to the passengers, the more expensive the ticket price would be. Thus, in order to normalize the travel distance with the effect of travel class, in present study, D_c measure is suggested as combination of distance and class of travel. Consequently, the “D_c” variable is calculated as follows

D_{c} = Distance * ClassRate

(2)

On the other hand, one of the problems of the RFM model is that it cannot distinguish the new customers from the regular ones (Reinartz and Kumar, 2000). The length variable means: “The days from the first visit date to the last visit date” (Reinartz and Kumar, 2000). Failure of considering the length (L) of the customer’s relationship with the organization caused problems in results of regarding the customer value. Such problems can induce more failures mainly in transportation industry where receiving services is not permanent. Thus, in most studies conducted on customer value, the length variable is also used along with other variables.

At the end of this section, it is worth mentioning that due to the specific characteristics of the transportation industry, LDcFR model is more comprehensive than the other models.

Clustering

Determining the variables L, Dc, F and R for each passenger cannot making sense for the respective airline company. Airlines are always seeking to identify groups of target customers for whom they can provide appropriate services. To achieve this goal, customer clustering has been introduced as effective and commonly used technique. As stated previously, the most popular and widely used algorithm for clustering process is K-means. Due to K-means problems, nowadays, this algorithm is used along with evolutionary algorithms. Since ICA is one of the fastest and most accurate evolutionary algorithms, it will be used for the purpose of passenger clustering in the present study.

Value extraction

Although RFM model has a great deal with determining the value of customers, this model works based on customers’ past transactions. RFM model cannot be a useful model in determining the value of customers unless the customer’s future behaviors are similar to their past behaviors. Accordingly, in order to determine the prospective valued customers, their future must be specified.

One of the most widely used methods in predicting the future behaviors of customers is Markov chain. Markov chain is a discrete random spatial process with a memoryless property. Memoryless characteristic of the Markov chain is indicative of the fact that: the prediction of the future states merely depends on the current states of the system and not on the past sequence of events and states. In what follows, the procedures of using Markov chain method in predicting variables of distance (D_c), frequency (F), Recency (R) and length (L) are explained.

Passenger’s Past transactional data are clustered with K-MICA algorithm, and cluster centers are determined.

The transition matrix is made with the placement of the cluster centers as transitions in matrix.

The transition matrix is formed:

3-1. All matrix elements are made zero.

3-2. Passenger’s new transactional data are clustered.

3-3. For each change in the status of each passenger in the new clusters compared to the old clusters, one unit of value is added to the corresponding element in the matrix.

The matrix is normalized in order to determine transition probabilities.

The future state of LDcFR variables is be predicted for each passenger.

The future value of each passengers is calculated based on the weighted RFM method (equation (3)).

value = w_{1} * {L + w}_{d} * {D + w}_{f} * {F + w}_{f} * R

(3)

In equation (3), the w₁, w₂, w₃ and w₄ are the weights of each of variables that are calculated using techniques such as analytic hierarchy process (AHP) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), etc.

The expected value of each customer is calculated using the equation (4).

LTV = \sum_{t = 1}^{T} P * value * (\frac{1}{1 + δ})^{t}

(4)

In equation (4), “value” refers to the weighted RFM value, “P” represents the transition probabilities, “δ” is the yearly discount rate and “t” is the horizon (in years).

Based on above descriptions, the final framework is shown in Figure 1.

Figure 1.

Final framework.

Case study

The proposed methodology was applied to the case of southwest airline’s passenger. The application of the methodology and results are summarized in this section.

Data description

The respective dataset consisted of 37 felids and 32,157 transaction records from 1369 Southwest airline’s passengers in 2011 and 2012.

Data selection and preprocessing

Among the features found in the dataset, only the properties associated with LDcFR model were needed to be used. Thus, only “passenger id,” “source,” “destination,” “distance,” “date of purchase ticket” and “flight class” fields were selected. An example of the projected data from the dataset is shown in Table 3.

Table 3.

An example of the selected data.

Passenger_Id	Source	Destination	Distance	Year	Month	Class
124	Midland/Odessa	Baltimore	1516	2012	05	F
775	Arizona	Oklahoma	833	2012	01	L

From this dataset, passengers’ transaction data in 2012 were separated and determined as the original data in calculating passengers’ value process and passengers’ transaction data in 2011 were used in building the transition matrix.

Data preparation

According to the examples shown in Table 3, the LDcFR model variables must be derived as shown in Table 4.

Table 4.

Calculating method model variables.

Variable	Calculation method	Formula
Length (L)	Time elapsed since the first purchase to the last one.	$Max (year . month) - Min (year . month)$
Distance (D_c)	Applied class impact to each distance	$Sum (Distance) : \forall Id$
Distance (D_c)	Total distance per ID
Frequency (F)	Appearance of each ID number in database	$Count ({Passenger}_{Id}) : \forall Id$
Recency (R)	Time elapsed since the customers’ last purchase to the present	$2013.01 - (year . month)$

In calculating the Recency, since the time interval was defined between January 2012 and December 2012, the January 2013 was considered as the present date, and Recency value per customer was calculated by subtracting this date from the last purchase.

Using the formula presented in Table 4, the variables of the LDcFR model per passenger were formed. An example of the data from the new dataset (Table 3) is shown in Table 5.

Table 5.

An example of final dataset.

Passenger_Id	Length (month)	Distance (class)	Frequency	Recency (month)
124	23	41,797	32	1
775	22	4257	5	2

Clustering process

The purpose of this section is to demonstrate the clustering process with K-MICA methods and explain the results.

Knowing the number of clusters in the clustering process is very important. Determining the number of clusters is possible with both methods, namely, expert opinion and statistical indicators. In this study, to determine the exact number of clusters, the Silhouette index¹ (statistical index) was used. The results of calculating Silhouette index for 3 to 10 clusters are illustrated in Figure 2. As shown in Figure 2, the index value in four clusters is greater than other numbers, thus the optimal number of clusters was set to 4.

Figure 2.

Comparison of the Silhouette width for different number of clusters.

After determining the optimal number of clusters, passengers’ data in the database were clustered using the ICA. Clustering results are brought in Table 6.

Table 6.

The centers of clusters.

Clusters	Length	Distance	Frequency	Recency	Number of passenger (cluster size)
Cluster 1	18	160,063.2	114	1	34
Cluster 2	23	56,166.8	43	12	106
Cluster 3	22	20,399.82	17	2	265
Cluster 4	13	2543.587	2	5	811

Expected value

According to the final research model (Figure 1), the next step is the prediction of passenger’s future state using Markov chain algorithm. In making the transition matrices, the results of clustering transactions in 2012 (Table 6) were considered as the new conditions, and the results of clustering customer transactions in 2011 were considered as the initial (past) conditions. In lieu of a change in status of any of the passengers on the created clusters, one unit was added to the corresponding element of the transition matrix. The resulted transition matrix is shown in Table 7.

Table 7.

The transition matrix.

Clusters	Cluster 1	Cluster 2	Cluster 3	Cluster 4	Churn
Cluster 1	31	6	0	0	0
Cluster 2	3	92	56	0	0
Cluster 3	0	8	168	85	1
Cluster 4	0	0	41	594	152

For instance, Table 7 indicates that three passengers who were in cluster 2 of the clustering transactions in 2011 have shifted to cluster 1 of the clustering transactions in 2012. With normalized results are shown in Table 7, the transition probabilities will be determined. The normalized results are shown in Figure 3.

Figure 3.

The transition probabilities.

According to equation (2), in order to calculate the value of each passenger, the weight of each of the variables was needed to be determined. Therefore, the AHP technique was utilized. Table 8 shows the results of AHP implementation.

Table 8.

Variable weight based on AHP.

	Length	Distance	Frequency	Recency	W
Length	1	0.3412	0.2648	1	0.114
Distance	2.9301	1	0.5296	3.1776	0.311
Frequency	3.7763	1.8881	1	3.1776	0.456
Recency	1	0.3146	0.3146	1	0.12

As shown in Table 8, from the perspective of airlines managers, the “Frequency” is the most valuable factor, and the “Distance” is the second important factor.

Practical implementation

After determining the transition probabilities and the weight of each variable, the future value of each passenger was estimated according to equations (2) and (3). Then, the future values of the passengers in each cluster were averaged as a basis for the value of each cluster. The results cluster ranking based on averaged values are brought in Table 9.

Table 9.

Ranking the clusters.

Clusters	Number of passenger	Average value	Rank
Cluster 1	33	0.57960	1
Cluster 2	106	0.25340	2
Cluster 3	265	0.18085	3
Cluster 4	811	0.09891	4

As shown, 33 passengers in cluster 1 are among the most valuable airline passengers for whom guarding strategies should be designed to protect them. This cluster of passengers will be highly profitable in the future. Passengers in clusters 2 and 3 are relatively valued, thus the company should pay more attention to them in order to increase their loyalty and profitability. More attention to these passengers can improve the airline’s position in the competitive market.

Validating the model

As shown in Table 8, according to expert’s opinion, the main factor in determining the passenger value is “Frequency” and then “Distance,” “Length” and “Recency,” respectively. Accordingly, with reference to the analysis of cluster 1, that is, cluster with the most valued passengers; it is clear that in this cluster, the situations of “Frequency,” “Distance” and “Recency” are much better than those of the other clusters (L↓D↑F↑R↑). On the other hand, the analysis of cluster 4, as the disloyal customers, illustrates that the situations of “Frequency,” “Distance” and “Length” variables are less likely than those of the other clusters (L↓D↓F↓R↑). In Table 10, the formed cluster analysis using statistical indicators is shown. In the last column of this table, the overall pattern of clusters which is the general results of the statistical indicators is shown.

Table 10.

Clusters pattern.

Clusters	Pattern based on the average	Pattern based on the median	General pattern
Cluster 1	L↓D↑F↑R↑	L↓D↑F↑R↑	L↓D↑F↑R↑
Cluster 2	L↑D↓F↓R↓	L↑D↑F↑R↓	L↑D↑F↑R↓
Cluster 3	L↑D↓F↓R↑	L↑D↓F↓R↑	L↑D↓F↓R↑
Cluster 4	L↓D↓F↓R↑	L↓D↓F↓R↓	L↓D↓F↓R↑

According to the patterns shown in Table 10, it can be concluded that the expert opinions regarding the value of the passengers have been approved in clusters results, which verify the proposed model.

Conclusion

The main purpose of this paper was to propose a comprehensive model to determine the value of customers in transportation industry. The value-prediction model developed in this study was made based on the combination of four variables, namely, “Length,” “Frequency,” “Distance” and “Recency” that has been named as LDcFR model. Unlike the other RFM-based models, in LDcFR, the “monetary value” is not directly used; because of dynamicity and instability in price of the ticket sales.

Although RFM model is one of the most well known and popular models in determining the customers’ value, using this model independently cannot be of great help to industry. This is due to the tendency of companies to develop limited strategies. Accordingly, in this paper, the ICA was used in clustering process of passengers. The rationales behind using ICA have been its high speed and accuracy.

Although RFM model has a great deal of power in determining the value of customers, this model works based on customers’ past transactions. Accordingly, in order to determine the prospective valued customers, their future behaviors must also be taken into account. Therefore, Markov chain was employed in the present study to predict the future transactions of the passengers. The expected value of each customer was calculated based on this prediction. Validity assessment results demonstrate the accuracy and correctness of developed method.

The aggregated results of the study are indicative of the fact that LDcFR model enjoys a high degree of accuracy. In addition to its accuracy, this model has higher speed compared to the other models, which is due to using ICA.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Note

Author Biographies

MohammadReza DehghaniZadeh received his MS degree in IT engineering at School of Industrial Engineering at the Iran University of Science and Technology (IUST), Tehran. He has contributed to several research papers in national and international conferences and entrepreneurship communities. His current research interests are knowledge discovery and data mining, customer relationship management, distributed systems and decision making theory. He is also interested in both the technical and business sides of entrepreneurship and startups.

Mohammad Fathian is professor of the School of Industrial Engineering of Iran University of Science and Technology, Tehran. He received his MS and PhD degrees in Industrial Engineering from the same university. Dr. Fathian is working in the areas of information technology, e-commerce and knowledge management. He has more than 60 journal papers and five books in the areas of industrial engineering and information technology.

MohammadReza Gholamian is an assistant professor in School of Industrial Engineering at the Iran University of Science and Technology (IUST), Tehran. He received his MS degree in Industrial Engineering from Isfahan University of Technology (IUT), Isfahan in 1998 and obtained PhD in Industrial Engineering from Amirkabir University of Technology (AUT), Tehran in 2005 for the work in the field of Hybrid Intelligent Decision Making Systems. Presently he is faculty member of IT Group in School of Industrial Engineering and is actively engaged in conducting Academic, Research and Development Programs in the field of Industrial Engineering and Information Technology. He has contributed more than 122 research papers to many national and international journals and conferences. Besides this, he has published 4 books by reputed publishers. His research interests include Data Mining, Soft Computing and Decision Theory.

References

Abbasi E, Owlia MS, Alizadeh S, et al. (2010) Defining marketing strategies by data mining for “Internet Sale System of Train Ticket”. In: The fourth Iran data mining conference. 31 October 2010. Tehran: IDMC.

Akhondzadeh-Noughabi

Alizadeh

Ahmadvand

A-M

et al. (2013) FTiS: A new model for effective urban management: A case study of urban systems in Iran. Cities 31: 394–403.

Alderighi

Cento

Piga

(2011) A case study of pricing strategies in European airline markets: The London – Amsterdam route. Journal of Air Transport Management 17(6): 369–373.

Atashpaz-Gargari E and Lucas C (2007) Imperialist competitive algorithm: an algorithm for optimization inspired by imperialistic competition. In: IEEE congress on evolutionary computation, CEC 2007, Singapore, 25–28 September 2007, pp.4661–4667. New York, NY: IEEE.

Berry

MJA

Linoff

(2004) Data Mining Techniques Second Edition-for Marketing, Sales, and Customer Relationship Management, Indianapolis, IN: Wiley.

Bisignani G (2010) Wings of change conference, Santiago. Available at: http://www.iata.org/pressroom/speeches/Pages/2010-03-24-01.aspx (accessed 31 May 2014).

Bohling

Bowman

LaValle

et al. (2006) CRM implementation effectiveness issues and insights. Journal of Service Research 9(2): 184–194.

Chang

E-C

Huang

S-C

H-H

(2010) Using K-means method and spectral clustering technique in an outfitter’s value analysis. Quality & Quantity 44(4): 807–815.

Chen

Y-S

Cheng

C-H

Lai

C-J

et al. (2012) Identifying patients in target customer segments using a two-stage clustering-classification approach: A hospital-based assessment. Computers in Biology and Medicine 42(2): 213–221.

10.

Cheng

C-H

Chen

Y-S

(2009) Classifying the segmentation of customer value via RFM model and RS theory. Expert Systems with Applications 36(3): 4176–4184.

11.

Cheng

L-C

Sun

L-M

(2012) Exploring consumer adoption of new services by analyzing the behavior of 3G subscribers: An empirical case study. Electronic Commerce Research and Applications 11(2): 89–100.

12.

Chiang

W-Y

(2012) Applying a new model of customer value on International Air Passengers’ Market in Taiwan. International Journal of Tourism Research 14(2): 116–123.

13.

Chiang

W-Y

(2014) Applying data mining with a new model on customer relationship management systems: A case of airline industry in Taiwan. Transportation Letters 6(2): 89–97.

14.

Fedorco

Hospodka

(2013) Airline pricing strategies in European Airline Market. Faculty of Transportation Sciences 2(8): 33–41.

15.

Gargari

Hashemzadeh

Rajabioun

et al. (2008) Colonial competitive algorithm: A novel approach for PID controller design in MIMO distillation column process. International Journal of Intelligent Computing and Cybernetics 1(3): 337–355.

16.

GTS

Lee

CKM

et al. (2012) Customer grouping for better resources allocation using GA based clustering technique. Expert Systems with Applications 39(2): 1979–1987.

17.

Y-H

Huang

TC-K

Kao

Y-H

(2013) Knowledge discovery of weighted RFM sequential patterns from customer sequence databases. Journal of Systems and Software 86(3): 779–788.

18.

Huang S-C, Wei J-T, Lin S-Y, et al. (2012) Using Bayesian network and LRFM model in a pediatric dental clinic. In: 2012 International symposium on computer, consumer and control (IS3C), Taichung, 4–6 June 2012, pp.20–23. New York, NY: IEEE.

19.

Hughes

(2006) Strategic Database Marketing, New York, NY: McGraw-Hill (first published in 1994).

20.

Kaveh

Talatahari

(2010) Optimum design of skeletal structures using imperialist competitive algorithm. Computers & Structures 88(21): 1220–1229.

21.

Khajvand

Tarokh

(2011) Estimating customer future value of different customer segments based on adapted RFM model in retail banking context. Procedia Computer Science 3: 1327–1332.

22.

KNEWS (2013) The cost of airline tickets is a function of demand and supply. Available at: http://www.kaieteurnewsonline.com/2013/07/10/the-cost-of-airline-tickets-is-a-function-of-demand-and-supply/ (accessed 25 February 2014).

23.

D-C

Dai

W-L

Tseng

W-T

(2011) A two-stage clustering method to analyze customer characteristics to build discriminative customer management: A case of textile manufacturing business. Expert Systems with Applications 38(6): 7186–7191.

24.

Lindenmeier

Tscheulin

(2008) The effects of inventory control and denied boarding on customer satisfaction: The case of capacity-based airline revenue management. Tourism Management 29(1): 32–43.

25.

Ling

Yen

(2001) Customer relationship management: An analysis framework and implementation strategies. Journal of Computer Information Systems 41(3): 82–97.

26.

Ngai

EWT

Xiu

Chau

DCK

(2009) Application of data mining techniques in customer relationship management: A literature review and classification. Expert Systems with Applications 36(2): 2592–2602.

27.

Niknam

Amiri

(2010) An efficient hybrid approach based on PSO, ACO and k-means for cluster analysis. Applied Soft Computing 10(1): 183–197.

28.

Niknam

Fard

Pourjafarian

et al. (2011) An efficient hybrid algorithm based on modified imperialist competitive algorithm and K-means for data clustering. Engineering Applications of Artificial Intelligence 24(2): 306–317.

29.

Rajabioun R, Hashemzadeh F, Atashpaz-Gargari E, et al. (2008) Identification of a MIMO evaporator and its decentralized PID controller tuning using colonial competitive algorithm. In: Be presented in IFAC World Congress.

30.

Reinartz

Kumar

(2000) On the profitability of long-life customers in a noncontractual setting: An empirical investigation and implications for marketing. Journal of Marketing 64(4): 17–35.

31.

Roshanaei M, Atashpaz-Gargari E and Lucas C (2008) Adaptive beamforming using colonial competitive algorithm. In: 2nd International joint conference on computational engineering, Vancouver, Canada.

32.

Sharma

Iyer

(2007) Country effects on CRM success. Journal of Relationship Marketing 5(4): 63–78.

33.

Tabaei Z and Fathian M (2011) Developing W-RFM model for customer value: An electronic retailing case study. In: 2011 3rd International conference on data mining and Intelligent Information Technology Applications (ICMiA), Macao, 24–26 October 2011, pp.304–307. New York, NY: IEEE.

34.

Wong

J-Y

Chung

P-H

(2007) Managing valuable Taiwanese airline passengers using knowledge discovery in database techniques. Journal of Air Transport Management 13(6): 362–370.

35.

Wei

J-T

Lin

S-Y

Weng

C-C

et al. (2012) A case study of applying LRFM model in market segmentation of a children’s dental clinic. Expert Systems with Applications 39(5): 5529–5533.

36.

Yeh

Yang

K-J

Ting

T-M

(2009) Knowledge discovery on RFM model using Bernoulli sequence. Expert Systems with Applications 36(3): 5866–5871.

37.

Zhongding Z, Xuemei M and Guangcan L (2009) Customer segmentation algorithm of wireless content service based on ant K-means. In: IFCSTA’09. International forum on computer science-technology and applications, vol. 1, Chongqing, 25–27 December 2009, pp.267–269. New York, NY: IEEE.