Spatiotemporal dynamics of geotourism resilience: Machine learning methodologies and key insights

The tourism industry, along with its various sub-sectors, offers numerous economic, social, and cultural benefits to host communities, particularly in rural areas. However, few studies have explored the prediction of geotourism resilience at the Likert scale using machine learning techniques, especially those focusing on feature selection through the 10-fold cross-validation technique. This study aims to predict geotourism resilience using several machine learning algorithms—including artificial neural networks (ANN), Bayes net (BN), logistic regression (LR), naive Bayes (NB), naive Bayes tree (NBTree), and random forest (RF)—in Quri-Qaleh village, Ravansar county, Kermanshah province, Iran. Data were collected from a randomly selected sample of 150 individuals through a structured questionnaire. The most important variables on geotourism resilience were selected based on the average merit of the information gain ratio (AM IGR ) feature selection technique. To evaluate the models' goodness-of-fit and predictive performance, several statistical metrics—including precision, recall, F 1-measure, area under the curve (AUC)—as well as the non-parametric Friedman test, were applied. The findings revealed 65.3% of respondents exhibited low to very low resilience, 10.7% moderate resilience, and 24% high to very high resilience. Among the predictive variables, product innovation (PI; AM IGR = 0.254), government support (GS; AM IGR = 0.253), increased destination attractiveness (IDA; AM IGR = 0.213), monthly income (MI; AM IGR = 0.203) and social capital (SC; AM IGR = 0.189) emerged as the five most influential variables. Comparative analysis indicated the NB algorithm outperformed other models in predicting geotourism resilience (mean of AUC = 0.826; F 1-measure = 0.627; recall = 0.622, and precision = 0.651). This research can assist decision-makers in environmental and rural development in identifying strategies to strengthen the resilience of geotourism. • We used 150 samples and 22 variables to classify geotourism resilience at 5-levels Likert scale. • Product innovation (PI) was the most important factor affecting geotourism resilience. • The NB outperformed the other ML algorithms in geotourism resilience prediction. • The local community of Quri Qala was severely disrupted during the COVID-19. • This study is applicable in identifying strategies to strengthen the resilience of geotourism.