En este trabajo se desarrolla los factores que influyen en la predicción de la Probabilidad de default (PD) en una entidad financiera, el principal objetivo del trabajo es encontrar perfiles relevantes que perfilen la PD y para ello se recurre al modelo de árbol de decisiones. Se analizaron aproximadamente 206 mil contratos de una cartera en específico de la entidad financiera y es con esta población que se realiza el modelo, la variable a predecir es el tipo de contrato (bueno, malo) presentando un desbalance en las categorías, buenos representa el 90% y malos el 10%. El modelo elegido fue aquel que necesito de realizar un balanceo de clases; los resultados obtenidos fue un modelo con un accuracy del 72% y un F1 score de 65% para los contratos malos, el árbol presenta una profundidad de 7, proporcionando granularidad en los perfiles hallados. Finalmente se detectaron 3 variables i...

This article presents the application of the non-parametric Random Forest method through supervised learning, as an extension of classification trees. The Random Forest algorithm arises as the grouping of several classification trees. Basically it randomly selects a number of variables with which each individual tree is constructed and predictions are made with these variables that will later be weighted through the calculation of the most voted class of these trees that were generated, to finally do the prediction by Random Forest. For the application, we worked with 3168 recorded voices, for which the results of an acoustic analysis are presented, registering variables such as frequency, spectrum, modulation, among others, seeking to obtain a pattern of identification and classification according to gender through a voice identifier. The data record used is in open access and can be do...

This article presents the application of the non-parametric Random Forest method through supervised learning, as an extension of classification trees. The Random Forest algorithm arises as the grouping of several classification trees. Basically it randomly selects a number of variables with which each individual tree is constructed and predictions are made with these variables that will later be weighted through the calculation of the most voted class of these trees that were generated, to finally do the prediction by Random Forest. For the application, we worked with 3168 recorded voices, for which the results of an acoustic analysis are presented, registering variables such as frequency, spectrum, modulation, among others, seeking to obtain a pattern of identification and classification according to gender through a voice identifier. The data record used is in open access and can be do...

This article presents the application of the non-parametric Random Forest method through supervised learning, as an extension of classification trees. The Random Forest algorithm arises as the grouping of several classification trees. Basically it randomly selects a number of variables with which each individual tree is constructed and predictions are made with these variables that will later be weighted through the calculation of the most voted class of these trees that were generated, to finally do the prediction by Random Forest. For the application, we worked with 3168 recorded voices, for which the results of an acoustic analysis are presented, registering variables such as frequency, spectrum, modulation, among others, seeking to obtain a pattern of identification and classification according to gender through a voice identifier. The data record used is in open access and can be do...

Este estudio se centra en el desarrollo de modelos de regresión utilizando técnicas avanzadas de aprendizaje automático para predecir el valor comercial de inmuebles en el contexto financiero peruano. Se exploran diversos modelos, como Random Forest, Support Vector Regression (SVR), Redes Neuronales, y XGBoost, siendo este último el que demostró mayor precisión al obtener un Error Porcentual Medio Absoluto de 17.89% y, por otra parte, una mayor flexibilidad al permitir controlar el comportamiento que debe tener cada variable independiente respecto a la variable objetivo. Además, se destaca la importancia de incluir variables como el área de construcción del bien inmueble y precio del metro cuadrado considerando su ubicación. Los resultados de este estudio proporcionan a las entidades financieras una herramienta robusta y eficiente para optimizar el proceso de tasación de inmue...