Application of machine learning to predict visitors’ green behavior in marine protected areas: evidence from Cyprus

Figures & data

Figure 1. Asymmetrical and configurational model of the study.

Table 1. Respondents’ profile (n = 332).

Characteristics	Frequency	Percentage
Gender
Male	140	42.2
Female	192	57.5
Age
18–29	26	7.8
30–49	104	31.3
50–64	127	38.3
Over 65	75	22.6
Education level
No schooling completed	36	10.5
Some high school	98	29.5
Associate degree/diploma	95	28.6
Trade/technical/Vocational training	64	19.3
Bachelor’s degree	29	8.7
Graduate and higher degree	10	3
Marital status
Married	206	62
Single	126	38
Income
Less than 1000 USD	72	21.7
1000 − 2999 USD	83	24.4
3000 − 5999 USD	152	45.8
More than 6000 USD	25	7.5
Nationality
British	115	34.6
Cypriot	30	8.7
German	27	8.1
Swedish	22	6.6
Norwegian	21	6.3
Turkish	18	5.4
Others	99	29.8
Total	332	100%

Table 2. Result of CFA, CR, AVE, MSV and α.

Variable				SFL	CR	AVE	MSV	α
Obsessive Passion (Marsh et al., 2013; Vallerand et al., 2003)					.914	.64	.068	.800
OP3	My environmental activities are the only thing that really turns me on.			.827
OP4	If I could, I would only do environmentally friendly activities.			.849
OP6	I have the impression that my passion toward the environment controls me.			.797
OP2	I have almost an obsessive feeling for the environment.			.787
OP1	I have difficulties controlling my urge to do environmentally friendly activities.			.769
OP5	My environmental activities are so exciting that I sometimes lose control over them.			.769
Harmonious Passion (Marsh et al., 2013; Vallerand et al., 2003)					.888	.57	.068	.755
HP2	The new things that I discover in the environment allow me to appreciate it even more.			.819
HP6	Taking care of the environment is in harmony with other things that are part of me.			.767
HP4	Helping the environment allows me to live a variety of experiences.			.803
HP3	Behaving environmentally friendly reflect the qualities I like about myself.			.766
HP5	My environmental activates are well integrated in my life.			.734
HP1	My environmentally friendly practices are in harmony with the other activities in my life.			.626
Pro-environmental Behaviour (Smith-Sebasto & D'Costa, 1995; Su et al., 2018; Thapa, 2010)					.879	.548	.044	.741
PEB3	When I see garbage, tree branches etc., I will put them in the trash bin.			.803
PEB5	I try to convince partners to protect the natural environment.			.786
PEB4	If there are environment cleaning activities, I am willing to attend.			.769
PEB2	I report to the destination administration on any environmental pollution or destruction.			.713
PEB1	I comply with relevant regulations to not destroy the environment.			.696
PEB6	I try to not disrupt the fauna and flora during my visit.			.666
Hedonism (Kim & Ritchie, 2014)					.883	.654	.391	.809
Hed3	Really enjoyed this tourism experience			.799
Hed2	Indulged in the activities			.914
Hed4	Exciting			.711
Hed1	Thrilled about having a new experience			.799
Knowledge (Kim & Ritchie, 2014)					.924	.802	.034	.895
Kno2	Knowledge			.928
Kno1	Exploratory			.887
Kno3	New culture			.871
Novelty(Kim & Ritchie, 2014)					.924	.527	.025	.726
Nov2	Unique			.736
Nov3	Different from previous experiences			.803
Nov4	Experienced something new			.757
Nov1	Once-in-a lifetime experience			.588
Refreshment (Kim & Ritchie, 2014)					.81	.523	.034	.723
Ref2	Enjoyed sense of freedom			.826
Ref1	Liberating			.777
Ref3	Refreshing			.735
Ref4	Revitalized			.515
Involvement (Kim & Ritchie, 2014)					.905	.763	.03	.873
Inv2	I enjoyed activities which I really wanted to do			.963
Inv3	I was interested in the main activities of this tourism experience			.882
Inv1	I visited a place where I really wanted to go			.762
Local culture (Kim & Ritchie, 2014)					.892	.735	.087	.857
Loc3	Local people in a destination were friendly			.944
Loc1	Good impressions about the local people			.828
Loc2	Closely experienced the local culture			.793
Meaningfulness (Kim & Ritchie, 2014)					.805	.583	.391	.764
Mng2	I did something important			.833
Mng1	I did something meaningful			.815
Mng3	Learned about myself			.626

Note: SFL: standardized factor loading; SFL is significant at the .001 level; AVE: average variance extracted; MSV: maximum shared squared variance; CR: composite reliability; α: Chronbach’s Alpha; hp = Harmonious passion; op = Obsessive Passion; peb = Pro-environmental behaviour; hed = Hedonism; kno = Knowledge; inv = Involvement; loc = Local culture, mng = Meaningfulness; nov = Novelty; ref = Refreshment.

Table 3. Cross-Tabulation Analysis of Obsessive Passion* Pro-environmental Behaviour.

			Peb							Total
			1	2	3	4	5	6	7
Op	1	Count	0	0	0	2	2	0	0	4
		% within op	0.0%	0.0%	0.0%	50.0%	50.0%	0.0%	0.0%	100.0%
		% within peb	0.0%	0.0%	0.0%	3.3%	1.6%	0.0%	0.0%	1.2%
		% of Total	0.0%	0.0%	0.0%	0.6%	0.6%	0.0%	0.0%	1.2%
	2	Count	0	2	9	6	12	6	1	36
		% within op	0.0%	5.6%	25.0%	16.7%	33.3%	16.7%	2.8%	100.0%
		% within peb	0.0%	15.4%	17.6%	10.0%	9.7%	7.4%	100.0%	10.8%
		% of Total	0.0%	0.6%	2.7%	1.8%	3.6%	1.8%	0.3%	10.8%
	3	Count	1	3	15	18	37	12	0	86
		% within op	1.2%	3.5%	17.4%	20.9%	43.0%	14.0%	0.0%	100.0%
		% within peb	50.0%	23.1%	29.4%	30.0%	29.8%	14.8%	0.0%	25.9%
		% of Total	0.3%	0.9%	4.5%	5.4%	11.1%	3.6%	0.0%	25.9%
	4	Count	1	3	17	23	44	22	0	110
		% within op	0.9%	2.7%	15.5%	20.9%	40.0%	20.0%	0.0%	100.0%
		% within peb	50.0%	23.1%	33.3%	38.3%	35.5%	27.2%	0.0%	33.1%
		% of Total	0.3%	0.9%	5.1%	6.9%	13.3%	6.6%	0.0%	33.1%
	5	Count	0	3	7	11	24	26	0	71
		% within op	0.0%	4.2%	9.9%	15.5%	33.8%	36.6%	0.0%	100.0%
		% within peb	0.0%	23.1%	13.7%	18.3%	19.4%	32.1%	0.0%	21.4%
		% of Total	0.0%	0.9%	2.1%	3.3%	7.2%	7.8%	0.0%	21.4%
	6	Count	0	2	2	0	5	14	0	23
		% within op	0.0%	8.7%	8.7%	0.0%	21.7%	60.9%	0.0%	100.0%
		% within peb	0.0%	15.4%	3.9%	0.0%	4.0%	17.3%	0.0%	6.9%
		% of Total	0.0%	0.6%	0.6%	0.0%	1.5%	4.2%	0.0%	6.9%
	7	Count	0	0	1	0	0	1	0	2
		% within op	0.0%	0.0%	50.0%	0.0%	0.0%	50.0%	0.0%	100.0%
		% within peb	0.0%	0.0%	2.0%	0.0%	0.0%	1.2%	0.0%	0.6%
		% of Total	0.0%	0.0%	0.3%	0.0%	0.0%	0.3%	0.0%	0.6%
Total		Count	2	13	51	60	124	81	1	332
		% within op	0.6%	3.9%	15.4%	18.1%	37.3%	24.4%	0.3%	100.0%
		% within peb	100.0%	100.0%	100.0%	100.0%	100.0%	100.0%	100.0%	100.0%
		% of Total	0.6%	3.9%	15.4%	18.1%	37.3%	24.4%	0.3%	100.0%

Note: peb = pro-environmental behavior, op = obsessive passion; Cramer’s V = .354, phi = 1.938, p < 0.000 indicating association between variables; Bold marks indicates 70 cases for ∼ Obssessive Passion → pro-environmental behavior and 15 cases for Obssessive Passion → ∼pro-environmental behavior; 1 represent Strongly disagree, 7 represents agree.

Table 4. Configural Models of high and low scores of Peb (model A, B and their negations).

A. peb = f(hed, kno, inv, loc, mng, nov, ref, gen,mar,age,inc,edu)					∼A. ∼peb = f(hed, kno, inv, loc, mng, nov, ref, gen,mar,age,inc,edu)
Models for predicting high score of peb	RC	UC	C		Models for predicting low score of peb	RC	UC	C
M1. age*gen*mar*inc*∼edu*hed*mng*loc*∼nov*kno*ref	.16	.09	.98		M1. age*∼gen*mar*inc*∼edu*hed*mng*loc*nov*inv*∼kno*ref	.21	.01	.91
M2. age*∼gen*mar*inc*∼edu*hed*mng*loc*nov*inv*∼kno*ref	.10	.01	1		M2. age*gen*mar*inc*∼edu*hed*mng*loc*∼nov*∼inv*kno*ref	.22	.06	.88
M3. age*∼gen*∼mar*inc*edu*hed*mng*loc*∼nov*inv*kno*ref	.09	.01	.99		M3. age*∼gen*∼mar*inc*edu*hed*mng*loc*∼nov*inv*kno*ref	.19	.02	.91
M4. age*∼gen*mar*inc*edu*hed*mng*loc*nov*inv*kno*∼ref	.11	.02	.98		M4. age*∼gen*mar*inc*edu*hed*mng*loc*nov*inv*kno*∼ref	.22	.02	.86
solution coverage: .23					solution coverage: .33
solution consistency: .98					solution consistency: .76
B. peb = f(hp, op, gen,mar,age,inc,edu)					∼B. ∼ peb = f(hp, op, gen,mar,age,inc,edu)
Models for predicting high score of peb	RC	UC	C		Models for predicting low score of peb	RC	UC	C
M1. op*hp*age*inc	.50	.09	.93		M1. ∼hp*∼op*age*∼gen*edu	.36	.01	.70
M2. ∼op*∼hp*age*∼gen*edu	.21	.01	.95		M2. ∼hp*∼op*age*∼gen*mar*inc	.29	.02	.73
M3. ∼op*∼hp*age*mar*inc	.29	.01	.94		M3. ∼hp*∼op*age*∼mar*inc*edu	.29	.007	.74
M4. ∼op*∼hp*age*inc*edu	.31	.01	.95		M4. hp*op*age*∼gen*∼mar*inc	.23	.01	.74
M5. op*hp*age*gen*mar	.32	.04	.94		M5. hp*op*age*∼mar*inc*edu	.28	.009	.72
M6. ∼op*∼hp*age*∼gen*∼mar*∼inc	.13	.01	.94		M6. hp*op*age*gen*mar*∼inc*edu	.29	.13	.80
M7. ∼op*∼hp*gen*mar*inc*∼edu	.19	.01	.96
solution coverage: .68					solution coverage: .64
solution consistency: .91					solution consistency: .60

Note: RC: raw coverage; C: consistency; UC: unique coverage; *: and; ∼ : negation; Peb: Pro-environmental behaviour; age : respondents’ age; gen: gender; mar: marital status; inc: income; edu: education; hed: Hedonism; kno: Knowledge; inv: Involvement; loc: Local culture, mng: Meaningfulness; nov: Novelty; ref: Refreshment; hp: harmonious passion; op: obsessive passion; marital status and gender are dummy variables: 0 indicates: single and men, while 1 indicates: married, women respectively.

Table 5. Configural models of high and low scores of Peb with all the antecedents (model C and its negation).

C. peb : f(hed, kno, inv, loc, mng, nov, ref, hp, op, gen, mar, age, inc, edu)	RC	UC	C
Models for predicting high scores Peb
M1. age*∼gen*mar*inc*∼edu*hed*mng*loc*nov*inv*∼kno*ref*∼op*∼hp	.09	.01	1
M2. age*∼gen*mar*inc*edu*hed*mng*loc*nov*inv*kno*∼ref*op*hp	.10	.02	.98
M3. age*gen*mar*inc*∼edu*hed*mng*loc*∼nov*inv*kno*ref*op*hp	.14	.07	.99
solution coverage: 0.19
solution consistency: 0.98
∼C. ∼ peb : f(hed, kno, inv, loc, mng, nov, ref, hp, op, gen, mar, age, inc, edu)	RC	UC	C
Models for predicting low scores Peb
M1. age*∼gen*mar*inc*∼edu*hed*mng*loc*nov*inv*∼kno*ref*∼op*∼hp	0.21	0.01	0.92
M2. age*∼gen*mar*inc*edu*hed*mng*loc*nov*inv*kno*∼ref*op*hp	0.22	0.02	0.87
solution coverage: 0.24
solution consistency: 0.84

Note: RC : raw coverage; C : consistency; UC: unique coverage; ∼ : negation; Peb: Pro-environmental behaviour; age: respondents’ age; gen: gender; mar: marital status; inc: income; edu: education; hed: Hedonism; kno: Knowledge; inv: Involvement; loc: Local culture, mng: Meaningfulness; nov: Novelty; ref: Refreshment; hp: harmonious passion; op: obsessive passion; marital status and gender are dummy variables: 0 indicates: single and men, while 1 indicates: married, women respectively.

Table 6. Evidence of predictive validity on two subsamples.

Configural model	Raw coverage		Consistency
age*∼gen*mar*inc*∼edu*hed*mng*loc *nov*inv*∼kno*ref*∼op*∼hp	0.09	1	.09	.01	1
Test of the same model with subsample I	Test of the same model with subsample II

Note: The fuzzy XY plot unveils the asymmetric relationship of the causal model and provides the predictive validity.

Table 7. ANFIS sensitivity analysis results

Model	Absent	Input	Output	RMSE
Model 1	–	Hed, Mng, Loc, Nov, Inv, Kno, Ref, Hp, Op	Peb	2.0922
Model 2	Hed	Mng, Loc, Nov, Inv, Kno, Ref, Hp, Op	Peb	5.5279
Model 3	Mng	Hed, Loc, Nov, Inv, Kno, Ref, Hp, Op	Peb	6.6864
Model 4	Loc	Hed, Mng, Nov, Inv, Kno, Ref, Hp, Op	Peb	5.0278
Model 5	Nov	Hed, Mng, Loc, Inv, Kno, Ref, Hp, Op	Peb	3.3427
Model 6	Inv	Hed, Mng, Loc, Nov, Kno, Ref, Hp, Op	Peb	5.7577
Model 7	Kno	Hed, Mng, Loc, Nov, Inv, Ref, Hp, Op	Peb	2.9652
Model 8	Ref	Hed, Mng, Loc, Nov, Inv, Kno, Hp, Op	Peb	3.7893
Model 9	Op	Hed, Mng, Loc, Nov, Inv, Kno, Ref, Hp	Peb	3.4022
Model 10	Hp	Hed, Mng, Loc, Nov, Inv, Kno, Ref, Op	Peb	3.4223

Note: RMSE = root mean square error.

Figure 2. ANFIS model structure.

Figure 3b. Surface view of the highest and lowest scored antecedents with Peb.

Table A.1. Results of EFA, and descriptive statistics of scale items (n = 332)

Factor	λ	Mean	Std. Deviation			Skewness		Kurtosis
				Eigenvalues	% of Variance	Statistic	Std. Error	Statistic	Std. Error
				5.865	13.963
OP3	.792	3.98	1.767			.004	.134	−1.056	.267
OP4	.822	4.05	1.763			.006	.134	−1.055	.267
OP6	.841	4.11	1.825			−.014	.134	−1.110	.267
OP2	.795	4.03	1.798			.003	.134	−1.067	.267
OP1	.682	3.95	1.832			.051	.134	−1.128	.267
OP5	.832	4.11	1.691			.009	.134	−.890	.267
				4.213	10.030
HP2	.810	4.24	1.310			−.058	.134	−.269	.267
HP6	.812	4.37	1.362			−.143	.134	−.241	.267
HP4	.769	4.29	1.369			−.063	.134	−.386	.267
HP3	.815	4.27	1.422			.045	.134	−.563	.267
HP5	.869	4.31	1.406			.078	.134	−.350	.267
HP1	.938	4.29	1.384			−.029	.134	−.316	.267
				3.588	8.543
PEB3	.906	4.98	1.288			−.533	.134	−.020	.267
PEB5	.909	4.99	1.424			−.540	.134	−.239	.267
PEB4	.931	5.04	1.339			−.423	.134	−.546	.267
PEB2	.909	4.86	1.569			−.321	.134	−.937	.267
PEB1	.875	5.07	1.284			−.432	.134	−.224	.267
PEB6	.863	5.23	1.541			−.788	.134	−.200	.267
				3.387	8.063
Hed3	.923	4.02	.851			−.951	.134	1.198	.267
Hed2	.791	3.71	.986			−.485	.134	−.288	.267
Hed4	.839	3.81	1.012			−.615	.134	−.253	.267
Hed1	.701	3.76	.997			−.791	.134	.291	.267
				2.973	7.079
Kno2	.744	3.49	1.070			−.408	.134	−.581	.267
Kno1	.838	3.42	1.047			−.323	.134	−.566	.267
Kno3	.824	2.90	1.027			.076	.134	−.450	.267
				2.813	6.698
Nov2	.775	3.91	1.052			−.878	.134	.252	.267
Nov3	.813	3.56	1.209			−.443	.134	−.746	.267
Nov4	.824	3.85	1.093			−.768	.134	−.152	.267
Nov1	.850	3.24	.817			−.035	.134	.210	.267
				2.247	5.350
Ref2	.848	3.26	.800			.141	.134	.241	.267
Ref1	.788	3.31	.878			.107	.134	.015	.267
Ref3	.838	3.29	.952			−.016	.134	−.379	.267
Ref4	.759	3.59	1.408			−.623	.134	−.996	.267
				2.114	5.034
Inv2	.771	3.55	1.432			−.589	.134	−1.060	.267
Inv3	.826	3.58	1.358			−.605	.134	−.945	.267
Inv1	.798	3.76	1.383			−.989	.134	−.329	.267
				1.801	4.289
Loc3	.804	3.80	1.286			−.999	.134	−.119	.267
Loc1	.735	3.87	1.304			−1.146	.134	.161	.267
Loc2	.818	3.36	.924			−1.380	.134	.891	.267
				1.139	2.713
Mng2	.837	3.45	.910			−1.000	.134	.250	.267
Mng1	.816	3.57	1.065			−.824	.134	.119	.267
Mng3	.649	3.34	.907			−.303	.134	−.225	.267

Note: λ is factor loading coefficient; Std. Deviation is standard deviation. pebi = Pro-environmental behaviour intention; hp = Harmonious passion; op = Obsessive passion; hed = Hedonism; kno = Knowledge; inv = Involvement; loc = Local culture, mng = Meaningfulness; nov = Novelty; ref = Refreshment; All items gauged by 5-point Likert scale ranging from strongly disagree (1) to strongly agree (5) except peb, hp and op that are ranging from Strongly disagree (1) to Strongly agree (7). Kaiser-Meyer-Olkin (KMO): .793; Bartlett's Test of Sphericity: Approx. Chi-Square: 8169.053, df: 861; Sig.: 0.000; All Eigen-values > 1; Skewness and Kurtosis provide evidence for normality.

Table A2. ANFIS best configuration for Peb according to membership function type and its category.

		Category of Membership Function
	Member Function	2	3
RMSE Values	trimf	5.5107	2.2728
	trapmf	11.2922	9.4735
	gbellmf	7.3679	2.1002
	gussmf	5.2425	2.0922
	guss2mf	17.0251	3.0516
	pimf	39.6398	11.6474
	desigmf	47.9745	3.9697
	psigmf	47.9745	3.9697

Note: Abbreviations of member functions: Difference between two sigmoid functions membership function (dsigmf); gaussian combination membership function (gauss2mf); Gaussian curve membership function (gaussmf); generalized bell membership function (gbellmf); p-shaped membership function (pimf); product of two sigmoidal membership function (psigmf); trapezoidal membership function (trapmf); triangular membership function (trimf).

Table A3. Result of discriminant validity analysis.

	1	2	3	4	5	6	7	8	9	10
(1) Obsessive passion	0.800
(2) Harmonious passion	0.260***	0.755
(3) Pro-environmental Behaviour	0.163**	0.209**	0.741
(4) Hedonism	0.123*	0.183**	0.076	0.809
(5) Knowledge	0.080	0.027	−0.041	0.125*	0.895
(6) Novelty	0.022	0.000	0.018	0.041	−0.122	0.726
(7) Refreshment	−0.060	−0.065	−0.135*	0.048	−0.184**	−0.159*	0.723
(8) Involvement	0.043	−0.084	−0.012	−0.037	0.174**	0.022	−0.139*	0.873
(9) Local culture	−0.002	0.027	0.041	0.294***	0.014	0.042	−0.119	0.022	0.857
(10) Meaningfulness	−0.012	0.082	0.061	0.625***	0.009	0.046	0.043	0.111	0.201**	0.764

Note: The non-diagonal elements are the correlations of the constructs along with their p-value indication while the diagonal elements (in bold) are the square roots of the Average Variance Extracted (AVE);.

*** p < 0.001.

** p < 0.010.

* p < 0.050.

Vallerand, R. J., Blanchard, C., Mageau, G. A., Koestner, R., Ratelle, C., Léonard, M., Gagné, M., & Marsolais, J. (2003). Les passions de l'ame: on obsessive and harmonious passion. Journal of Personality and Social Psychology, 85(4), 756–767. https://doi.org/10.1037/0022-3514.85.4.756

 PubMed Web of Science ®Google Scholar

Smith-Sebasto, N. J., & D'Costa, A. (1995). Designing a Likert-type scale to predict environmentally responsible behavior in undergraduate students: A multistep process. The Journal of Environmental Education, 27(1), 14–20. https://doi.org/10.1080/00958964.1995.9941967

 Google Scholar

Kim, J. H., & Ritchie, J. B. (2014). Cross-cultural validation of a memorable tourism experience scale (MTES). Journal of Travel Research, 53(3), 323–335. https://doi.org/10.1177/0047287513496468

 Web of Science ®Google Scholar