Maestría en Analítica Aplicada · 2026

Predicción de caudales del río Bogotá mediante modelos estadísticos y de machine learning

This project presents a hydrological forecasting model for the Bogotá River section at Puente La Virgen (Cota, Cundinamarca), aimed at anticipating streamflow using statistical methods and machine learning techniques. Historical discharge data (2007–2022) at this location and precipitation data from a nearby station were integrated. The study implemented traditional models (ARIMA/ARIMAX), machine learning algorithms (HistGradientBoosting, LSTM), and hybrid approaches, evaluated with statistical metrics (R², RMSE, MAE) and hydrological indicators (Nash-Sutcliffe Efficiency, NSE). The results indicate that hybrid models outperform individual approaches, reaching acceptable values of R² = 0.84 during training and R² = 0.77 during testing. Furthermore, these models show improved capabilities in detecting extreme streamflows, defined by percentile thresholds (90% and 95%). These findings demonstrate that combining statistical and computational methods strengthens peak flow prediction and provides a solid foundation for an early warning system in flood risk management within the Bogotá Savanna.