Plataforma móvil para el monitoreo y analítica predictiva de datos microclimáticos: caso de estudio torre meteorológica ubicada en la Facultad de Ingeniería en Sistemas, Electrónica e Industrial de la Universidad Técnica de Ambato

Abstract

This research aimed to develop a mobile platform for monitoring and predicting meteorological variables to determine microclimates on the campus of the Faculty of Systems Engineering, Electronics, and Industrial Engineering. The study was structured in three phases. First, raw data from the institutional weather tower was collected and cleaned by applying exploratory data analysis, average signal filtering, and temporal feature engineering to generate optimal numerical matrices for temperature, solar radiation, and barometric pressure. Second, four predictive algorithms —Random Forest, XGBoost, Stacked LSTM, and CNN-LSTM— were trained and evaluated using R², MAE, and RMSE metrics. Results demonstrated that no single universal model is optimal: Random Forest achieved the highest accuracy for temperature (R² = 0.88), while XGBoost outperformed the rest for solar radiation (R² = 0.82) and barometric pressure (R² = 0.91). Additionally, a heuristic rainfall probability estimator based on pressure and temperature deltas was implemented due to the absence of a physical rain gauge. Third, the complete technological ecosystem was built: an asynchronous FastAPI backend following Clean Architecture principles integrating serialized models, and a cross-platform React Native mobile application featuring real-time dashboard, historical visualization, and four-hour forecast modules. The platform was validated through integration and usability testing, confirming its operability and low latency in production environments.