Use of climate data and machine learning for rainfall prediction in the sustainable management of ecotourism in Francisco Beltrão (PR, Brazil)

Abstract

The use of climate data and artificial intelligence techniques has emerged as a promising strategy for developing tools aimed at planning and sustainable management of activities in natural environments. This study developed a predictive model for rainfall occurrence in Francisco Beltrão, Paraná, Brazil) based on historical records of meteorological variables such as temperature, relative humidity, evaporation, solar radiation, and precipitation. The dataset, composed of daily observations from an official weather station, underwent preprocessing, temporal feature extraction, and binary categorization of rainfall volumes. A Random Forest model was trained and evaluated using performance metrics including accuracy, precision, recall, F1-score, and ROC AUC. Results indicated 83% accuracy in predicting rainy and dry days, with 91% of dry days correctly identified. The most influential variables were evaporation and relative humidity, confirming known climate dynamics. Graphical analyses contributed to understanding model behavior under class imbalance and highlighted variable importance. The proposed system showed practical potential to support decision-making in trail scheduling, visitor safety, and area management within ecotourism contexts, enhancing sustainable land use and environmental conservation. The model's structure is replicable in other regions and can be expanded with real-time forecasting data to increase operational effectiveness.