Modelo de predição da desordem do colapso das colônias por mineração de dados

Abstract

Bees are essential pillars of biodiversity and global food security. However, the phenomenon known as Colony Collapse Disorder (CCD), driven by both biological and anthropogenic factors, has decimated populations and caused severe damage to biodiversity. In this context, Precision Beekeeping (PB) emerges as a strategic response, leveraging the Internet of Things (IoT) to monitor beehives in an automated manner and transform physical variables into actionable data. This dissertation presents the development of a PB prototype specifically designed for the stingless bee species Jataí (Tetragonisca angustula). The system integrates temperature, humidity, and mass sensors with a single-board computer, the Raspberry Pi Zero 2W, which transmits data to a cloud platform, enabling remote monitoring via a dashboard. To ensure data accuracy, the Savitzky-Golay filter was employed, a smoothing technique that removes experimental noise without distorting the underlying trends of time series. The theoretical foundation was established through a systematic literature mapping, identifying gaps in computational applications to apiculture. The methodology involved data mining processes and the application of Machine Learning (ML) algorithms. Decision Tree, Random Forest, and Gradient Boosting models were compared to determine the best performance in predicting anomalies associated with colony collapse. The results confirm the technical feasibility of integrating loT and ML for the management of INPA model hives. In addition to deepening the understanding of Jataí bee behavior, this work validates a computational architecture capable of supporting preventive decision-making. In summary, this research provides a low-cost, high-precision tool, essential for optimizing apicultural management and strengthening the sustainability of bee populations in the face of modern environmental threats.