Examinando por Materia "Modelos predictivos"
Mostrando 1 - 8 de 8
Resultados por página
Opciones de ordenación
Ítem A predictive approach based on fuzzy cognitive maps with federated learning(Universidad EAFIT, 2023) Garatejo Vargas, Edison Camilo; Aguilar Castro, José Lizandro; Hoyos, WilliamÍtem Analítica de datos aplicada a la cobranza de cartera(Universidad EAFIT, 2019) Montoya Yepes, Juan David; Parra Giraldo, Diana CarolinaIn order to improve the portfolio collection work performed daily at Cobroactivo LLC, it was decided to use the data stored in this company to optimize its processes. To do this, it was necessary to implement data analytics models, creating a Big Data environment that would allow efficient access to information through a Data Warehouse, as well as some tools to make an exploratory analysis of the existing data and evaluate the weak pointsthat must be improved when managing users. Warehouse, as well as some tools to make an exploratory analysis of the existing data and evaluate the weak points that must be improved when managing users. Likewise, three Machine Learning models were developed in charge of debugging the debtors, predicting the payment probabilities and optimally recommending which advisor should be assigned to each debtor and which is the appropriate contact channel for him / her. Finally, two web applications were developed. The first allows the monitoring of the company's internal processes by automating repetitive processes and decreasing their execution time from weeks o seconds; the second allows the monitoring of collection work by customers and banks, thereby offering added value.Ítem Aplicación de modelos de inteligencia artificial y aprendizaje automático para la previsión de precios y la optimización de portafolios : un enfoque integrado con datos estructurados y no estructurados con el fin de compararse con el S&P 500 como benchmark(Universidad EAFIT, 2023) Vélez García, Santiago; Botero Ramírez, Juan CarlosThis study presents an integrated approach of artificial intelligence and machine learning models, combining neural networks for price forecasting and portfolio optimization in the financial industry. The results show that the integrated approach outperforms other financial analysis methods and provides more effective tools for market professionals compared to a buy and hold strategy represented in the analysis by the S&P500. The artificial intelligence and machine learning models used in this study enable the identification of patterns and trends in financial data, helping investors make more informed and accurate decisions. Furthermore, the study demonstrates that the inclusion of unstructured data, such as news and social networks, in financial analysis can significantly improve the accuracy of price predictions achieving an R2greater than 65% and portfolio optimization.Ítem Contribuciones desde un modelo predictivo para identificar el perfil de riesgo del defraudador interno en una entidad financiera de Colombia(Universidad EAFIT, 2018-05-29) Ruíz Galeano, Fabio Hernán; Torres Guerra, Idier AlbeiroFinancial services companies are a fundamental part of the country's economy, and their continued growth has made them more susceptible to fraud due to the lack of prevention and controls that mitigate the threats that go hand in hand with technological innovation, new products and consumption characteristics -- This problem generates negative economic and social impacts for both the company and the country -- That is why it was identified the need to develop the risk profile of the internal fraudster in a financial entity, which allows the frauds prevention and detection from the characterization of employees involved in fraudulent actions, to achieve this, the mixed investigation was applied, which allowed the collection of data on fraud materialized by employees, their numerical measurement and respective statistical analysis to test the hypothesis, as well as a literature review -- From the construction of the Random Forest model, and with the objective of finding employees with high probabilities of incurring in incorrect acts, the profile of the internal fraudster of the entity was identified, the employee who work in the branch network, who have atypical transactions and where the wage relationships Vs the discounts made to the employee do not keep a proportion -- Employees who hold operational positions which represent the 44.5% of the company, are the most likely to incur internal fraud, 66.4% of possible fraudsters are women; moreover, country’s regions that present greater alert are the central region with 43.6%, followed by Bogotá and Sabana with 28.7%Ítem Implementación de un modelo de scoring de crédito para Mexichem Colombia SAS(Universidad EAFIT, 2024) Vargas Izquierdo, Daniel Camilo; Rojas Ormaza, Brayan RicardoThis project sought to implement a credit scoring model for Mexichem Colombia SAS using machine learning techniques to predict the probability of default in companies. Four algorithms were compared: decision trees, random forest, gradient boosting and neural networks, each with unique characteristics in terms of accuracy and handling of complex data. The research included the selection and evaluation of relevant variables using the Gini index and recursive elimination techniques to avoid overfitting. The results helped to identify the most effective model to predict credit risks, optimizing financial decision-making.Ítem Machine Learning para la estimación del riesgo de crédito en una cartera de consumo(Universidad EAFIT, 2021) Ossa Giraldo, Wbeimar; Jaramillo Marin, Veronica; Rojas Ormaza, Brayan RicardoFinancial entities, due to their business nature, are inherently exposed to credit risk, for this reason, they are continually searching for new ways to measure the probability of default of clients requesting a loan. This research aims to comparing the precision of a logistic regression model against basic Machine Learning models for estimating credit risk in a consumer loan portfolio, these methodologies are emerging as a key tool for estimating risks due to their flexibility and learning capacity. For this, the Logistic Regression, Random Forest, Support Vector Machine and Multilayer Perceptron models were used, making a comparison in the efficiency of the estimation of the clients that are going to default, and obtaining as a result that the most balanced model at time of evaluation is the Random Forest.Ítem Segmentación de clientes y definición de alertas para la prevención de riesgos de lavado de activos y financiación del terrorismo (SARLAFT): un estudio económico aplicado a entidad financiera colombiana en 2017(Universidad EAFIT, 2017) Amaya Molina, Mateo; Chaparro Cardona, Juan CamiloThe XXI century governments get the challenge to fortress their internal structure and controls against financial crimes of which they can be object -- The money laundering and the terrorism financing are the most frequently committed financial crimes around the world -- This type of activities is threatening the economic equilibrium in silence way turn and it´s representing a risk while talking about international trade relationships too -- The informatics advances in the field of data mining and statistics created a new complete landscape for those governments that try to get preventing controls and to identify that kind of activities -- These controls are required to those type of private organizations operating in the country in depending of the precise risk to each economic sector they belong -- This research develops an innovative method composed by the combination of techniques of data mining applying and the economics sector analysis looking for a possible answer to the financial sector entities controlling designed by SARLAFT in the Colombian governmentÍtem Walk-forward Optimization Algorithm for Time-Series Models(Universidad EAFIT, 2022) Castro Marín, Carlos Andrés; Almonacid Hurtado, Paula María