Inteligencia de datos en acción: modelos predictivos aplicados a la toma de decisiones y campañas financieras en el sector bancario
Loading...
Official URL
Full text at PDC
Publication date
2025
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
En un mundo globalizado, y dentro de un ecosistema cuyo dinamismo y competitividad aumentan día a día, el volumen de datos disponibles crece bajo una tasa exponencial y alarmante. Un estudio reciente estima que a cada persona se le asignan más de 2 MB de datos por segundo, lo que impulsa la necesidad de desarrollar competencias avanzadas en análisis de datos (Rosário, Moniz, & Cruz, 2021).
Dado este contexto, es posible preguntarse ¿Tiene la ciencia de datos capacidad de modificar el paradigma actual? ¿Se puede cambiar la forma en la que las empresas, se anticipan y actúan en función al comportamiento de sus clientes? El presente informe busca articular dos sectores muy importantes del mercado: Data Science y la Inteligencia de negocios. Asimismo, el foco estará puesto en el área de marketing, aplicado a campañas de contacto telefónico para la promoción de productos bancarios. El objetivo es claro, mas no sencillo. Durante el desarrollo de este trabajo, se buscar predecir el comportamiento de los clientes a la hora de contratar productos financieros a partir de datos históricos de campañas. Para ello, se utiliza el conjunto de datos de campañas bancarias del UCI, una colección de bases de datos y generadores de datos ampliamente utilizados por la comunidad de machine learning para el análisis empírico de algoritmos (University of California, Irvine, n.d.), que contiene estadísticas sobre clientes y campañas telefónicas (Moro, Cortez, & Rita, 2014). El mismo se puede encontrar a través del siguiente link1. Con este fin, se emplean diversas técnicas de clasificación binaria y machine learning, considerando en todo momento los desafíos inherentes a tipo de análisis, como el desbalance de clases, el sobreajuste, la importancia de las variables y la optimización de parámetros para cada modelo.
Profundizando levemente en las metodologías utilizadas (dado que se describen en detalle más adelante), podemos encontrar la regresión logística básica, que servirá de punto inicial para las comparativas. Posteriormente se utilizan modelos más complejos como las Redes Neuronales, Random Forest y Bagging, Suport Vector Machine y algoritmos de Boosting (Catboost y XGBoost). Con la finalidad de seleccionar el modelo optimo, se evalúan y comparan todos los métodos en un ranking bajo un enfoque corporativo.
La comparación de este enfoque bajo métricas específicas del negocio, como AUC (Area Under the Curve) y la tasa de fallos, trasciende la precisión académica, maximizando su aplicabilidad real para la optimización de recursos y clasificación.
La aplicación de modelos de machine learning en marketing ha demostrado tener un impacto directo en la personalización de estrategias, la mejora de la eficiencia de campañas y el aumento de la rentabilidad de las decisiones. No se trata solamente de predecir, sino de optimizar recursos, segmentar inteligentemente y anticipar decisiones del consumidor (Wedel & Kannan, 2016). Así, la ciencia de datos no solo aporta herramientas técnicas, sino que redefine la lógica operativa del marketing actual.
Abstract: In a globalized world, and within an ecosystem whose dynamism and competitiveness increase day by day, the volume of available data grows at an exponential and alarming rate. A recent study estimates that each person is assigned more than 2 MB of data per second, driving the need to develop advanced competencies in data analysis (Rosário, Moniz, & Cruz, 2021). Given this context, one may ask: ¿Does data science have the capacity to shift the current paradigm? ¿Can it change the way companies anticipate and act based on customer behavior? This report aims to articulate two highly relevant sectors of the market: Data Science and Business Intelligence. Moreover, the focus is placed on marketing, specifically applied to telephone contact campaigns for the promotion of banking products. The objective is clear, though not simple. Throughout this work, the goal is to predict customer behavior when subscribing to financial products using historical campaign data. For this purpose, the UCI Bank Marketing dataset is utilized—a widely used collection of databases and data generators in the machine learning community for empirical algorithm analysis (University of California, Irvine, n.d.)—which contains statistics on clients and telephone campaigns (Moro, Cortez, & Rita, 2014). The dataset can be accessed through the following link.2 To achieve this, various binary classification and machine learning techniques are employed, always considering the inherent challenges of this type of analysis, such as class imbalance, overfitting, variable importance, and parameter optimization for each model. Briefly delving into the methodologies used (as they are described in detail later), one can find basic logistic regression, serving as a starting point for comparisons. Subsequently, more complex models are applied, including Neural Networks, Random Forest and Bagging, Support Vector Machines, and Boosting algorithms (CatBoost and XGBoost). To select the optimal model, all m methods are evaluated and ranked under a corporate-focused approach. Comparing this approach using business-specific metrics, such as AUC (Area Under the Curve) and failure rate, goes beyond academic precision, maximizing its real- world applicability for resource optimization and classification. The application of machine learning models in marketing has demonstrated a direct impact on strategy personalization, campaign efficiency, and profitability enhancement. It is not just about prediction, but about optimizing resources, intelligently segmenting, and anticipating consumer decisions (Wedel & Kannan, 2016). In this way, data science not only provides technical tools but also redefines the operational logic of modern marketing.
Abstract: In a globalized world, and within an ecosystem whose dynamism and competitiveness increase day by day, the volume of available data grows at an exponential and alarming rate. A recent study estimates that each person is assigned more than 2 MB of data per second, driving the need to develop advanced competencies in data analysis (Rosário, Moniz, & Cruz, 2021). Given this context, one may ask: ¿Does data science have the capacity to shift the current paradigm? ¿Can it change the way companies anticipate and act based on customer behavior? This report aims to articulate two highly relevant sectors of the market: Data Science and Business Intelligence. Moreover, the focus is placed on marketing, specifically applied to telephone contact campaigns for the promotion of banking products. The objective is clear, though not simple. Throughout this work, the goal is to predict customer behavior when subscribing to financial products using historical campaign data. For this purpose, the UCI Bank Marketing dataset is utilized—a widely used collection of databases and data generators in the machine learning community for empirical algorithm analysis (University of California, Irvine, n.d.)—which contains statistics on clients and telephone campaigns (Moro, Cortez, & Rita, 2014). The dataset can be accessed through the following link.2 To achieve this, various binary classification and machine learning techniques are employed, always considering the inherent challenges of this type of analysis, such as class imbalance, overfitting, variable importance, and parameter optimization for each model. Briefly delving into the methodologies used (as they are described in detail later), one can find basic logistic regression, serving as a starting point for comparisons. Subsequently, more complex models are applied, including Neural Networks, Random Forest and Bagging, Support Vector Machines, and Boosting algorithms (CatBoost and XGBoost). To select the optimal model, all m methods are evaluated and ranked under a corporate-focused approach. Comparing this approach using business-specific metrics, such as AUC (Area Under the Curve) and failure rate, goes beyond academic precision, maximizing its real- world applicability for resource optimization and classification. The application of machine learning models in marketing has demonstrated a direct impact on strategy personalization, campaign efficiency, and profitability enhancement. It is not just about prediction, but about optimizing resources, intelligently segmenting, and anticipating consumer decisions (Wedel & Kannan, 2016). In this way, data science not only provides technical tools but also redefines the operational logic of modern marketing.












