Examinando por Materia "KNIME"

Mostrando 1 - 2 de 2
  • No hay miniatura disponible
    Publicación
    Modelos de riesgo crediticio para reducir el riesgo de crédito en la cartera hipotecaria del Banco Unión
    (Universidad EAFIT, 2026-02-26) Córdoba Ruiz, Ana Patricia; Rojas Ormaza, Brayan Ricardo
    Banco Unión’s mortgage portfolio has increased its exposure to risk as a result of post-pandemic economic uncertainty, evidenced by higher unemployment levels, inflationary pressures, and sustained increases in interest rates. Within this context, the study examines credit risk models designed to enhance the predictive capacity of portfolio deterioration and to support decision-making in the management of mortgage portfolios. The analysis relies on a historical database provided by Banco Unión, which contains information on performing and impaired mortgage loans. This dataset includes macroeconomic variables, credit-specific attributes, and the sociodemographic and financial characteristics of borrowers. From a methodological perspective, a binary Logit model was estimated to assess the probability of default, analyzing the marginal impact of key factors such as income level, interest rate, macroeconomic conditions, and loan maturity. In addition, a Decision Tree model was implemented on the KNIME platform using AutoML techniques, following sample balancing procedures. The predictive performance of the models was evaluated using standard metrics in the financial sector: the area under the ROC curve (AUC), sensitivity (recall), and precision. The results show that the Logit model provides a strong explanatory framework for the determinants of credit risk, while the Decision Tree exhibits greater sensitivity in identifying impaired borrowers while maintaining high levels of precision. By combining both perspectives, Banco Unión gains a more robust analytical tool to anticipate portfolio deterioration, improve customer segmentation, and strengthen credit risk monitoring and mitigation strategies in a changing economic environment, thereby aligning statistical standards with operational and regulatory criteria.
  • No hay miniatura disponible
    Publicación
    Optimización de un portafolio de inversión con acciones del Colcap aplicando técnicas de machine learning
    (Universidad EAFIT, 2024) Osorio Buitrón, Maribel; Rico Villareal, Juan David; Rojas Ormanza, Bryan Ricardo
    Based on the increases in the volumes of historical information on the shares of public companies, the question arises as to whether it is possible to create better optimized portfolios than those generated from traditional theory, making use of recent innovations in artificial intelligence and analysis of data from the last decade. For this reason, the present research aims to compare the traditional theory of portfolio optimization with the recent data analysis methodologies applied to Colcap. The methodology used is based on twelve previous investigations which had tested and demonstrated the performance of different machine learning models on stock exchanges around the world, such as S&P500, NASDAQ, DAX, SET and Colcap. From here, the best prospects were selected and applied to Colcap shares, to predict the future movement in the share price based on the historical behavior of certain significant variables and then compared with the traditional methodology. It was found that the best prediction model in the price movement applied to Colcap is the Random Forest, and the variables that best explain the future changes in the price of the shares of this exchange are the closing price of the share, the TRM index and the Colcap index. In addition, machine learning models managed to optimize portfolios with a smaller number of shares and higher returns backed by historical information.