Aplicación de modelos de Aprendizaje Automático y Aprendizaje Profundo, para el análisis del dataset del BCE

Abstract

Abstract: The Central Bank of Ecuador (BCE) plays a fundamental role in managing international reserves, monitoring the liquidity of the financial system, and designing economic policies. However, most of the historical series generated by the Central Bank are not analyzed comprehensively; that is, they are not fully utilized to identify patterns and anticipate macroeconomic risks. This thesis analyzes the behavior of international reserves, the monetary aggregate (M2), and the GDP per capita rate in Ecuador as representative indicators of different areas of the Central Bank's operations. Each indicator is addressed as a central focus within its respective analytical objective, allowing for the study of external stability, the liquidity of the financial system, and the performance of the real sector of the economy. To this end, Key Development Methodology (KDD) is applied, using Machine Learning and Deep Learning techniques, including linear regression, Lasso regression, and convolutional neural networks (CNN-1D). These models are evaluated using MAE, RMSE, R², and directional accuracy metrics to ensure a fair comparison with models used in related studies. Finally, the models are interpreted, along with how the variables influence their predictions.