Exploring Apulia’s Regional Tourism Attractiveness through the Lens of Sustainability: A Machine Learning Approach and Counterfactual Explainability Process

Abstract

Visitor attraction dynamics lead tourism industry paths. A complex artificial neural network model was built to predict the incoming tourism flow in the Apulia region of Southern Italy as a function of the heterogeneity of the tourism supply available in this area. Open data from the Regional Tourism Observatory were targeted. Information on the distribution of facilities and activities that attract regional tourist flows was collected and grouped by municipality. An artificial neural network model was built with total tourist attendance as the dependent variable and tourist attractions as regressors. The Root Mean Square Error (RMSE) was used to select the optimal model using the lowest value. The final model was run with a hidden layer consisting of three neurons and a decay value of 0.01. A Multi-Objective Counterfactual model (MOC) was then constructed using a randomly selected row of normalized data frame to validate a useful tool in increasing total tourist attendance by 20% over that of the randomly selected municipality. A Garson’s variables importance plot indicated natural landscapes such as beaches, sea caves, and natural parks have a primary role expressed in terms of variable importance in the AI algorithm when used as an innovative methodology for evaluating tourism flows in the Apulia region. A further MOC model built using a randomly selected row of normalized data frame showed convents, libraries, historical buildings, public gardens, and museums as the top five features most modified to improve total attendance in a randomly selected municipality. Use of AI modeling revealed that the implementation of nature-based solutions may speed up the flow of tourism in the Apulia region while also promoting sustainable social development.

1. Introduction

Globally, tourism is a growing industry [1]. Beyond supporting economic convergence and expansion, tourism is an important enterprise for developing and non-developing nations [1]. In addition to education levels, price and income levels, the size of the economies of the countries sending and receiving tourists, cultural heritage and natural resources, tourism development is influenced by a plethora of economic, social and infrastructural factors, including transportation, geopolitical conditions, investment, culture, peace, environment, people and tourist flow [2].

From the perspective of the tourism sector growth, the concept of destination attractiveness is key and, microscopically speaking, is linked to the broader concept of territorial attractiveness, which is an expression of both the magnitude of inflows and the destination’s endowment of tangible and intangible resources and how individuals value a given area [3]. Practically speaking, an area may become more attractive if it has acquired more attractive factors or because other areas have lost some of theirs, acquiring a greater likelihood of being chosen by tourists.

In Southern Italy, and particularly in the Apulia region, tourism has appeared to be growing strongly in the past five years, despite the decline of the COVID-19 pandemic [4], mainly due to the attractive power of the coast, climate, historical and archaeological sites, and agricultural products, among others [5]. Its share of gross domestic product (GDP) has reached 13.6% with 9 billion euros. The tourism labor market and business structure have a strong social impact on the region, with nearly 36% of enterprises belonging to this sector (1 for every 12 people) and 135,000 workers. This important role has been guided by a strong strategic and management approach, such as two participatory 10-year strategic plans and three Destination Management Organization (DMO) agencies, including Pugliapromozione (https://www.agenziapugliapromozione.it/portal/ accessed on 15 May 2024).

Against this background, the present work has implemented an Artificial Intelligence (AI) methodology, called deep learning [6,7], to predict incoming tourism flows in the Apulia region, Southern Italy. It is well-known that AI is a growing field with a wide range of applications and research topics. Intelligent software is increasingly being used to automate ordinary work, analyze rumors or images, and identify relationships and patterns in social and economic issues. The use of machine learning processes in the more heterogeneous areas of managing and predicting economic behavior and trends represents a branch of applications of such techniques that is of great interest and rapidly expanding. This predictive approach is complemented by counterfactual methodology (Multi-Objective Model) [8] to overcome the “black box” related to the explainability processes of machine learning models.

The present research aimed to develop a complex artificial neural network model to answer the research question of how to predict the incoming tourist flow in the Apulia region as a function of the heterogeneity of the tourist offers present in the territory. We aimed to combine the strong territorial need to identify associations between the diverse and heterogeneous tourism supply and the greater or lesser tourist interest in light of the supply present in the territory related to the distribution of territorial accommodation facilities.

2. Methods

2.1. Regional Tourism Observatory Open Data

The data used to structure this study were open data freely available from the regional tourism observatory website and downloadable at the following link https://www.agenziapugliapromozione.it/portal/osservatorio-del-turismo accessed on 15 May 2024. The data under study are pertaining to the year 2022, depending on the available update.

2.2. Statistical Learning and Machine Learning Approach

The database under analysis was constructed through open data on the Apulia Region website, specifically in the Regional Tourism Observatory section. The data under analysis are related to the distribution of facilities and activities attracting the flow of tourists at the regional level. The database was constructed considering individual municipalities as statistical units. The analysis of municipal facilities was carried out by distinguishing between the provinces to which they belong. The database contains the number of tourist attractors per municipality. In order to assess the different distributions of them throughout the regional area, the entire sample is grouped based on the province they belong to, and the values expressed as mean and SD, and median (IQR) (

Table 1. Description of the Apulia region tourist attractions by province. N:253. All data are shown as mean ± sd, median (iqr) for continuous variables.

	Bari		BAT		Brindisi		Foggia		Lecce		Taranto
	Mean ± sd	Median (iqr)	Mean ± sd	Median (iqr)	Mean ± sd	Median (iqr)	Mean ± sd	Median (iqr)	Mean ± sd	Median (iqr)	Mean ± sd	Median (iqr)	p Value *
Arrivals (Italians)	19,657.179 ± 45,968.85	5043 (13,503)	13,336.9 ± 14,878.604	7794 (22,702.75)	19,195.25 ± 34,109.144	2254.5 (12,411.75)	15,773.531 ± 44,713.718	590 (10,099)	9571.613 ± 27,766.27	1096 (3626)	10,092.36 ± 16,969.652	3980 (6520)	0.01
Presence (Italians)	47,112.538 ± 101,890.674	12,643 (30,635)	30,087.5 ± 30,820.309	21,720 (53,060.25)	75,450.3 ± 150,983.973	5105.5 (29,378)	73,814.878 ± 251,342.06	2086 (30,618)	46,094.473 ± 134,534.164	3850 (15,849)	42,057.28 ± 70,965.375	11,412 (45,057)	0.11
Arrivals (Foreigners)	15,390.744 ± 46,875.192	2141 (4379)	4823 ± 6303.325	1811 (7775.75)	11,891.7 ± 23,217.18	1289 (6264.5)	3688.167 ± 11,962.284	101.5 (1173.5)	3967.129 ± 15,406.263	618 (1524)	2767.167 ± 5561.916	825 (1723.25)	<0.01
Presence (Foreigners)	39,689.128 ± 111,297.621	5453 (14,233.5)	11,887.7 ± 14,647.34	5265 (19,553)	41,654.5 ± 83,299.998	4290.5 (23,614.75)	15,229.729 ± 59,129.081	321 (3250)	13,759.355 ± 44,420.864	2183 (5513)	10,570.333 ± 21,066.718	2862 (6585.75)	<0.01
Total Arrivals	35,047.923 ± 92,629.25	6276 (19,289)	18,159.9 ± 21,078.901	9605 (31,790.25)	31,086.95 ± 56,615.943	3339.5 (17,837.75)	19,386.429 ± 56,382.53	736 (10,746)	13,538.742 ± 41,980.424	1823 (5036)	12,748.84 ± 21,606.276	4981 (9673)	<0.01
Total Presence	86,801.667 ± 212,281.091	18,096 (50,800.5)	41,975.2 ± 44,971.108	28,457 (72,298.75)	117,104.8 ± 227,448.143	9032.5 (52,065)	88,733.796 ± 309,019.542	2309 (33,324)	59,853.828 ± 173,111.707	6099 (24,898)	52,204.8 ± 86,719.743	13,883 (50,399)	0.04
Natural Areas	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0.05 ± 0.224	0 (0)	0.107 ± 0.366	0 (0)	0.052 ± 0.222	0 (0)	0.034 ± 0.186	0 (0)	0.69
Basilicas	0.22 ± 0.525	0 (0)	0.4 ± 0.843	0 (0)	0.35 ± 0.489	0 (1)	0.357 ± 0.483	0 (1)	0.144 ± 0.5	0 (0)	0.207 ± 0.491	0 (0)	<0.01
Hypogean Churches	0.439 ± 0.867	0 (1)	0.5 ± 0.707	0 (1)	1.1 ± 1.252	1 (1.25)	1.018 ± 1.314	1 (2)	0.433 ± 0.877	0 (1)	0.862 ± 1.505	0 (1)	<0.01
Caves	0.073 ± 0.264	0 (0)	0.1 ± 0.316	0 (0)	0 ± 0	0 (0)	0.018 ± 0.134	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.03
Museums	1.098 ± 1.96	1 (1)	1.8 ± 1.135	2 (1.5)	1.15 ± 1.599	1 (1.25)	0.839 ± 0.987	1 (1.25)	0.474 ± 1.191	0 (1)	0.655 ± 1.078	0 (1)	<0.01
Historic Buildings	0.878 ± 2.326	0 (1)	1.2 ± 1.989	0 (1.75)	0.85 ± 2.277	0 (1)	0.625 ± 1.244	0 (1)	0.536 ± 1.128	0 (1)	0.69 ± 1.775	0 (1)	0.94
Archaeological Parks	0.146 ± 0.478	0 (0)	1 ± 1.7	0 (1)	0.5 ± 0.688	0 (1)	0.464 ± 0.785	0 (1)	0.361 ± 0.616	0 (1)	0.414 ± 1.15	0 (0)	0.07
Towers	0.122 ± 0.331	0 (0)	0.2 ± 0.422	0 (0)	0.6 ± 1.095	0 (1)	0.25 ± 0.477	0 (0)	0.175 ± 0.677	0 (0)	0.069 ± 0.258	0 (0)	0.02
Churches	0.463 ± 0.674	0 (1)	0.8 ± 0.919	1 (1)	0.9 ± 1.586	0.5 (1)	0.482 ± 1.128	0 (0)	0.196 ± 0.716	0 (0)	0.241 ± 0.511	0 (0)	<0.01
Historical Theaters	0.268 ± 0.708	0 (0)	0.3 ± 0.483	0 (0.75)	0.3 ± 0.47	0 (1)	0.107 ± 0.312	0 (0)	0.052 ± 0.265	0 (0)	0.103 ± 0.31	0 (0)	<0.01
Entertainment Sites	0.488 ± 1.399	0 (0)	0.2 ± 0.422	0 (0)	0.3 ± 0.571	0 (0.25)	0.089 ± 0.345	0 (0)	0.113 ± 0.497	0 (0)	0.172 ± 0.602	0 (0)	0.09
Trulli	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.39
Nature Reserves	0.024 ± 0.156	0 (0)	0.2 ± 0.422	0 (0)	0.15 ± 0.366	0 (0)	0.196 ± 0.483	0 (0)	0.021 ± 0.143	0 (0)	0.138 ± 0.441	0 (0)	0.02
Gravines	0.195 ± 0.679	0 (0)	0.1 ± 0.316	0 (0)	0.15 ± 0.366	0 (0)	0.054 ± 0.227	0 (0)	0.041 ± 0.2	0 (0)	0.379 ± 0.862	0 (0)	0.02
Farmhouses	0.146 ± 0.358	0 (0)	0 ± 0	0 (0)	0.1 ± 0.308	0 (0)	0.018 ± 0.134	0 (0)	0.031 ± 0.174	0 (0)	0.138 ± 0.351	0 (0)	0.03
Beaches	0.707 ± 2.124	0 (0)	1 ± 1.886	0 (1)	0.65 ± 1.268	0 (0.25)	0.786 ± 2.462	0 (0.25)	0.412 ± 0.997	0 (0)	0.483 ± 0.738	0 (1)	0.57
Libraries	0.146 ± 0.422	0 (0)	0 ± 0	0 (0)	0.05 ± 0.224	0 (0)	0.089 ± 0.438	0 (0)	0.052 ± 0.335	0 (0)	0.034 ± 0.186	0 (0)	0.34
Romanesque Churches	0.415 ± 1.048	0 (0)	0.4 ± 0.699	0 (0.75)	0.6 ± 0.995	0 (1)	0.054 ± 0.227	0 (0)	0.247 ± 0.541	0 (0)	1.241 ± 2.278	0 (1)	<0.01
Convents	0.195 ± 0.459	0 (0)	0.1 ± 0.316	0 (0)	0.05 ± 0.224	0 (0)	0.232 ± 0.467	0 (0)	0.082 ± 0.312	0 (0)	0.172 ± 0.759	0 (0)	0.09
Landings	0.146 ± 0.478	0 (0)	0.1 ± 0.316	0 (0)	0.25 ± 0.55	0 (0)	0.036 ± 0.187	0 (0)	0.134 ± 0.513	0 (0)	0.172 ± 0.539	0 (0)	0.40
Castles	0.293 ± 0.68	0 (0)	0.4 ± 0.699	0 (0.75)	0.4 ± 0.598	0 (1)	0.268 ± 0.447	0 (1)	0.144 ± 0.353	0 (0)	0.241 ± 0.435	0 (0)	0.23
Nature Parks	0.024 ± 0.156	0 (0)	0.1 ± 0.316	0 (0)	0.1 ± 0.308	0 (0)	0.036 ± 0.187	0 (0)	0.041 ± 0.2	0 (0)	0.034 ± 0.186	0 (0)	0.64
Orthodox Worship Sites	0.073 ± 0.469	0 (0)	0.1 ± 0.316	0 (0)	0.05 ± 0.224	0 (0)	0.018 ± 0.134	0 (0)	0.021 ± 0.203	0 (0)	0 ± 0	0 (0)	0.38
Ports	0.195 ± 0.813	0 (0)	0.3 ± 0.483	0 (0.75)	0.2 ± 0.894	0 (0)	0.143 ± 0.401	0 (0)	0.072 ± 0.297	0 (0)	0.207 ± 1.114	0 (0)	0.13
Public Gardens	0.024 ± 0.156	0 (0)	0.2 ± 0.422	0 (0)	0.35 ± 0.489	0 (1)	0.107 ± 0.454	0 (0)	0.134 ± 0.471	0 (0)	0.138 ± 0.351	0 (0)	<0.01
Lakes	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.036 ± 0.187	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.21
Eco-museums	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.031 ± 0.174	0 (0)	0 ± 0	0 (0)	0.63
Amphitheaters	0 ± 0	0 (0)	0 ± 0	0 (0)	0.1 ± 0.308	0 (0)	0.018 ± 0.134	0 (0)	0.021 ± 0.203	0 (0)	0 ± 0	0 (0)	0.07
Marine Reserves	0 ± 0	0 (0)	0 ± 0	0 (0)	0.05 ± 0.224	0 (0)	0.018 ± 0.134	0 (0)	0.01 ± 0.102	0 (0)	0.034 ± 0.186	0 (0)	0.67
Sea Caves	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.143 ± 0.724	0 (0)	0.041 ± 0.2	0 (0)	0 ± 0	0 (0)	0.66
Mosques	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.39
WWF Oasis	0.073 ± 0.346	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.01 ± 0.102	0 (0)	0.034 ± 0.186	0 (0)	0.40
Hypogean Crushers	0 ± 0	0 (0)	0 ± 0	0 (0)	0.15 ± 0.366	0 (0)	0.018 ± 0.134	0 (0)	0.124 ± 0.331	0 (0)	0.103 ± 0.557	0 (0)	0.02
National Parks	0.024 ± 0.156	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.018 ± 0.134	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.64
Synagogues	0 ± 0	0 (0)	0.1 ± 0.316	0 (0)	0.05 ± 0.224	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	<0.01
Mills	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.018 ± 0.134	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.62
Sites of Catholic Worship	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0 ± 0	0 (0)	0.01 ± 0.102	0 (0)	0.034 ± 0.186	0 (0)	0.59

* Wilcoxon’s effect size for independent samples. Bold means significance.

).

Normal distributions of quantitative variables were tested using the Kolmogorov–Smirnov test. Therefore, data were reported as mean ± standard deviation (M ± SD) for continuous measures and frequency and percentages (%) for all categorical variables. A non-parametric approach was adopted to show any statistical differences between provinces, a p-value equal to or less than 0.05 was chosen to assess statistical differences (Kruskal–Wallis sum rank test). All the data were normalized using min–max normalization as follows (X − min(X))/(max(X) − min(X)).

The entire dataset was adopted to build an artificial neural network model on Total Tourist presence as the dependent variable and tourist attractions as regressors. The Caret package was used to build the neural network model. The model was trained using a hidden layer size of 3 to 9 with decay values of 0.001, 0.01, 0.1, and a linear activation function. The Root Mean Square Error (RMSE) was used to select the optimal model using the smallest value [9]. The final model was run with a hidden layer consisting of three neurons and a decay value of 0.01. Figure 1 shows the Apulia region point of interest percentage according to municipality. Figure 2 shows the architecture of the neural network. Figure 3, shows the variation of the RMSE value according to the combination of the hidden layer size and decay values. The relative importance of each variable as an absolute magnitude was evaluated using Garson’s algorithm (Figure 4). Olden’s algorithm [10] evaluated the relative importance of the input variables in the neural networks as the sum of the product of the raw weights of the input-hidden and hidden-output connections (Figure 5). The Counterfactuals package was adopted to build a Multi-Objective Counterfactual (MOC) [8] model using a randomly selected row of normalized data frames. The MOC model was constructed to increase total tourist attendance by 20% over that of the randomly selected municipality. The Multi-Object Counterfactual model was built considering “Beaches”, “WWF oasis”, “Caves”, “Sea Caves”, “Archaeological Parks”, ”Marine Reserves”, “Hypogean Churches”, “Castles” as fixed features (not modifiable), and epsilon of zero (no penalization) for 175 generations and an Individual Conditional Expectation (ICE) curve population initialization strategy [11]. All statistical analyses were performed using RStudio 2023.03.1, with packages: tidyverse, rstatix, gmodels, caret, kableExtra, epiR, iml, ggplot2, and counterfactuals.

3. Results

The description of the distribution of tourist attraction factors of the Apulia Region distinguished by province is shown in Table 1. The evaluation of the provinces shows a statistically significant difference between the incoming tourist flows. Italian arrivals turn out to be statistically different in the six provinces, in particular average values of 19,657.179 ± 45,968.85, 13,336.9 ± 14,878.604, 19,195.25 ± 34,109.144, 15,773.531 ± 44,713.718, 9571.613 ± 27,766.27, 10,092.36 ± 16,969.652 are shown relative to Bari, BAT, Brindisi, Foggia, Lecce and Taranto. At the same time, foreign arrivals are different in the provinces, with average values of 15,390.744 ± 46,875.192, 3688.167 ± 11,962.284, 3967.129 ± 15,406.263, 2767.167 ± 5561.916 relative to Bari, Barletta-Andria-Trani (BAT), Brindisi, Foggia, Lecce, and Taranto. This statistically significant difference is also present for the presence of foreign tourists (mean values: 39,689.128 ± 111,297.621, 11,887.7 ± 14,647.34, 41,654.5 ± 83,299.998, 15,229.729 ± 59,129.081, 13,759.355 ± 44,420.864, 10,570.333 ± 21,066.718 related to Bari, BAT, Brindisi, Foggia, Lecce, and Taranto) but no statistically significant differences are shown for the presence of distinct Italian tourists in Apulian provinces. The analysis of the presence of tourist attractions among the different provinces of Apulia showed statistically significant differences in both morphological and environmental and cultural component aspects. The attractiveness related to the environmental and morphological component appears to be statistically significantly distinct with respect to “Natural Areas,” “Nature Reserves”, and “Gravinas” and “Caves”. The attractiveness related to the environmental and morphological component seems to be statistically significant compared to “Natural Areas”, “Nature Reserves”, and “Gravines”. The average presence of natural areas dislocated among the provinces is 0.024 ± 0.156, 0 ± 0, 0.05 ± 0.224, 0.107 ± 0.36, 0.052 ± 0.222 and 0.034 ± 0.186, respectively, for Bari, BAT, Brindisi, Foggia, Lecce, and Taranto. Parallel to the natural areas, the natural reserves in the region are distributed differently in a statistically significant way, an average of 0.024 ± 0.156,0.2 ± 0.421, 0.15 ± 0.366, 0.196 ± 0.48, 0.021 ± 0.14 and 0.138 ± 0.441 for Bari, BAT, Brindisi, Foggia, Lecce and Taranto, respectively. The presence of tourist-attractive gravines is shown to be statistically significantly different among the provinces analyzed as an average of 0.195 ± 0.679, 0.1 ± 0.316, 0.15 ± 0.36, 0.054 ± 0.22, 0.041 ± 0.2 and 0.379 ± 0.862 for Bari, BAT, Brindisi, Foggia, Lecce, and Taranto, respectively. Attractors of a sociocultural nature that showed a different distribution over the territory at a statistically significant level include places of worship and historical value (Basilicas, Hypogean Churches, Romanesque Churches, Towers, Churches, Historical Theaters, Hypogean Crushers and Synagogues). Relative to the provinces of Bari, Bat, Brindisi, Foggia, Lecce and Taranto, the mean presence of the aforementioned points of interest are 0.22 ± 0.525, 0.4 ± 0.84, 0.35 ± 0.48, 0.357 ± 0.48, 0.144 ± 0.5 and 0.207 ± 0.491 for Basilicas, 0.439 ± 0.86, 0.5 ± 0.707, 1.1 ± 1.252, 1.018 ± 1.314, 0.433 ± 0.877 and 0.862 ± 1.50 for Hypogean Churches, 0.415 ± 1.048, 0.4 ± 0.699, 0.6 ± 0.995, 0.054 ± 0.227, 0.247 ± 0.541 and 1.241 ± 2.278 for Romanesque Churches, 0.122 ± 0.331, 0.2 ± 0.422, 0.6 ± 1.095, 0.25 ± 0.477, 0.175 ± 0.677 and 0.069 ± 0.258 for Towers, 0.463 ± 0.674, 0.8 ± 0.919, 0.9 ± 1.586, 0.482 ± 1.128 and 0.196 ± 0.716 for Churches, 0.268 ± 0.708, 0.3 ± 0.483, 0.3 ± 0.47, 0.107 ± 0.312 and 0.052 ± 0.265 for Historical Theaters, 0 ± 0, 0 ± 0, 0.15 ± 0.366, 0.018 ± 0.134, 0.124 ± 0.331 and 0.103 ± 0.557 for Hypogean Crushers, 0 ± 0, 0.1 ± 0.316, 0.05 ± 0.224, 0 ± 0, 0 ± 0 and 0 ± 0 for Synagogues. The distribution of Museums is statistically significantly different among the provinces in the territory, in particular the average attendance is 1.098 ± 1.96, 1.8 ± 1.135, 1.15 ± 1.599, 0.839 ± 0.987 and 0.474 ± 1.191 for Bari, BAT, Brindisi, Foggia, Lecce, and Taranto. Finally, agritourism supply is differentially located among the provinces with an average presence of 0.146 ± 0.358, 0 ± 0, 0.1 ± 0.308, 0.018 ± 0.134, 0.031 ± 0.174 and 0.138 ± 0.351. Figure 1 shows the Apulia region point of interest percentage according to municipality. Figure 2 shows the final Neural Network Model Architecture plot run with a hidden layer consisting of three neurons and a decay value of 0.01. Figure 3 shows the best hyperparameter tuning selection plot using the RMSE metric. Figure 4 and Figure 5 show Garson’s and Olden’s variable importance plots, respectively. Particularly, Figure 4 underlines the role of Beaches as the most associated points of interest to Total presence in the Apulia Region followed by Sea Caves, Natural Parks, and Convents. Figure 5 once again highlights the role of beaches and seaside behavior on Apulia’s tourist attractiveness. Figure 6 shows a bar plot of relative counterfactual changes and Figure 7 shows a counterfactual parallel plot, where real characteristics are shown in blue, while counterfactuals are shown in gray. Figure 6 shows Convents, Libraries, Historical Buildings, Public Gardens, and Museums as the first five most changed features to improving total presences in a randomly selected municipality.

4. Discussion

This research aimed to use a complex artificial neural network model to predict the incoming tourist flow in the Apulia region of Southern Italy as a function of the heterogeneity of the tourism supply available in this area. To achieve this goal, a supervised machine learning approach was implemented. This approach enables the evaluation of regional tourism attractiveness by dividing the analysis into two successive steps, that is, evaluating the recurring parametric and nonparametric relationships underlying the connection between tourist points of interest and overall tourist arrivals and, subsequently, evaluating the potential effect of changes in points of interest to promote a percentage increase in tourist arrivals according to a MOC model.

As the main finding, the evaluation of the role of different points of interest on regional tourism flow showed that natural landscape and characteristic sites such as beaches, sea caves and natural parks take a primary role expressed in terms of variable importance in the AI algorithm. Along these lines, comparing our results with the recent IPSOS report (www.ipsos.com accessed on 15 May 2024) on the attractiveness of Apulia in 2023, we found an overlap between the reports of respondents in the IPSOS survey and the findings of our AI algorithm. In fact, the IPSOS report identified beach facilities and access to the sea as the predominant factor in choosing a tourist destination among respondents, followed by natural and historical points of interest.

To explain these findings, it is worthwhile to keep in mind that, in terms of tourism resources, natural landscapes have an undeniable value and potential to attract tourists, as well as being a more sustainable choice for both visitors and the environment. Here, previous research has widely examined the restorative power of nature in treehouse hotels and shown that the physiological and psychological benefits of forest recreation and sleeping in the treetops have a positive influence on repeat and future visits [12,13]. Visitors who participate more frequently in nature-based recreational activities have been shown to gain more health resources than those who are less involved in nature during their leisure time. Nature-induced positive emotions have the potential to strengthen bonds within families and communities through shared experiences in parks, which in turn build social capital [14]. In addition, the emotional effect of overcoming challenges experienced in natural areas produces important benefits for the individual in terms of reducing self-destructive and antisocial behaviors [15], and improving self-esteem and self-confidence, which can also influence spiritual health. Apulia fits perfectly into this naturalistic scenario being a region rich in natural wonders. Whether traveling inland or along the coast, a vacation in Apulia allows visitors to experience breathtaking scenery and views: the “Saline” of Margherita di Savoia, the Gargano National Park and the Torre Guaceto Nature Reserve are just a few of them. When visitors reach natural destinations, human–nature interaction can induce restorative experiences such as direct attention recovery, physical stress relief and innate emotional affiliation [16,17]. These experiences are beneficial to the physical, psychological, spiritual, and psychosocial recovery of visitors. Therefore, the restorative outcomes of nature-based tourism encourage visitors to consider nature as a personal resource for health and well-being [18,19]. Not least, the preservation of biological diversity and ecosystem services are among the benefits of attractive tourism motivated by nature-based landscapes. Therefore, nature-based tourism, in addition to attracting more tourists, can fulfill the conservation mandate of natural areas while contributing to the sustainability of the area [20]. Some limitations of this study should be acknowledged. Of these, assessing tourism flows through the open data made available by the region could be a flaw, in conjunction with the failure to use data on tourism services offered and the pathway related to tourism management of individual micro-areas. At the least, future research should focus on collecting and integrating confounding and modulating flow data in a cross-sectional manner in order to develop more efficient and effective complex models.

5. Conclusions

Nature-based landscapes such as beaches, sea caves, and natural parks play a primary role in terms of important variables in the AI algorithm when implemented as an innovative methodology to assess tourism flows in the Apulia region, Southern Italy. Thus, nature-based solutions may boost tourism development as a source of both economy and sustainability and are also desirable for paths of transition toward a sustainable planet. The present research, beyond representing an innovative approach to the evaluation of territorial and tourism promotion activities, stands out as a starting point on which to lay the foundations of future implementation that could primarily involve the end user and the operator by advantageously modulating the flows of supply and demand.