An Explainable AI-Powered Carbon Prediction and Machine Learning for Carbon-Aware Tourism in the Community

Tourism contributes substantially to greenhouse gas emissions. This study developed a trip-level carbon prediction framework for community-based tourism, serving as a surrogate model for activity-based carbon accounting equations. To prevent target leakage, predictor features were restricted to 34 pre-travel proxies, excluding post-travel parameters. Five machine learning algorithms, including Multiple Linear Regression, Random Forest, eXtreme Gradient Boosting, Light Gradient-Boosting Machine, and Categorical Boosting, were evaluated using 13,000 synthetic trip profiles generated from 650 empirical baseline respondents (partitioned into 10,400 development scenarios from 520 respondents and 2600 hold-out scenarios from 130 respondents). LINE Mini App provided instantaneous carbon estimations and personalized recommendations. Categorical Boosting achieved predictive performance, with a mean absolute error of 1.28 kilograms of carbon dioxide equivalent (kgCO2e), root mean square error of 1.96 kgCO2e, mean absolute percentage error of 6.90%, and coefficient of determination of 0.948, outperforming linear baselines. Explainable Artificial Intelligence using SHapley Additive exPlanations indicated that the model captures primary pre-travel drivers, led by travel distance, stay duration, travel budget, and travel planning score. External field evaluation: 105 independent cohort users generated 1297 recommendation impressions, 688 clicks, and 347 acceptances, achieving a 53.05% click-through rate, a 50.44% acceptance rate, a 247.64 ms mean response time, and an 86.40% user feedback score, supporting proactive decision choice architecture for sustainable community-based tourism.