Aplicación de un modelo de regresión logística para predecir inconsistencias en la descarga de información durante el proceso de vinculación de clientes al servicio de energía eléctrica
No Thumbnail Available
Date
2024
Authors
Directors
Journal Title
Journal ISSN
Volume Title
Publisher
Institución Universitaria Pascual Bravo
Abstract
Para una empresa de servicios públicos domiciliarios la vinculación de clientes al servicio de energía eléctrica es un proceso vital en el negocio de la energía eléctrica por lo tanto es necesario garantizar la seguridad y calidad en este proceso. La verificación de solicitudes de conexión al servicio de energía eléctrica toma largos tiempos, es propensa a errores y genera altos costos operativos, además de insatisfacción entre los usuarios. Este proyecto propone el uso de herramientas de Machine Learning para automatizar la identificación de inconsistencias en los cobros de materiales y actividades y creando un modelo que permita predecir inconsistencias, mejorando la eficiencia y eliminando la aleatoriedad en las verificaciones, lo que optimizará recursos, reducirá quejas y fortalecerá la experiencia del cliente. El objetivo general es desarrollar un sistema predictivo que reduzca los tiempos manuales en un 90%, detecte inconsistencias de manera anticipada y permita estrategias de mejora continua. El método incluye la extracción y análisis de datos del sistema transaccional entre enero y octubre de 2024, para crear un modelo que identifique las solicitudes con posibles problemas. El balanceo y análisis de datos atípicos fueron de gran relevancia en el estudio. Aunque modelos como AdaBoostClassifier y XGBClassifier ofrecieron alta precisión, fueron descartados por problemas de sobreajuste. Se elige el modelo basado en Regresión Logistica con un accuracy de 86% ya que se asegura que no esta memorizando.
Abstract For a utility company, the connection of customers to the electric power service is a vital process in the electric power business, therefore it is necessary to guarantee the safety and quality in this process. The verification of requests for connection to the electric power service takes long times, is prone to errors and generates high operating costs, as well as dissatisfaction among users. This project proposes the use of Machine Learning tools to automate the identification of inconsistencies in the charges for materials and activities and creating a model that allows predicting inconsistencies, improving efficiency and eliminating randomness in verifications, which will optimize resources, reduce complaints and strengthen the customer experience. The general objective is to develop a predictive system that reduces manual times by 90%, detects inconsistencies in advance and allows continuous improvement strategies. The method includes the extraction and analysis of data from the transactional system between January and October 2024, to create a model that identifies requests with potential problems. Balancing and analysis of outliers were of great relevance in the study. Although models such as AdaBoostClassifier and XGBClassifier offered high accuracy, they were discarded due to overfitting problems. The model based on Logistic Regression with an accuracy of 86% is chosen since it is ensured that it is not memorizing.
Abstract For a utility company, the connection of customers to the electric power service is a vital process in the electric power business, therefore it is necessary to guarantee the safety and quality in this process. The verification of requests for connection to the electric power service takes long times, is prone to errors and generates high operating costs, as well as dissatisfaction among users. This project proposes the use of Machine Learning tools to automate the identification of inconsistencies in the charges for materials and activities and creating a model that allows predicting inconsistencies, improving efficiency and eliminating randomness in verifications, which will optimize resources, reduce complaints and strengthen the customer experience. The general objective is to develop a predictive system that reduces manual times by 90%, detects inconsistencies in advance and allows continuous improvement strategies. The method includes the extraction and analysis of data from the transactional system between January and October 2024, to create a model that identifies requests with potential problems. Balancing and analysis of outliers were of great relevance in the study. Although models such as AdaBoostClassifier and XGBClassifier offered high accuracy, they were discarded due to overfitting problems. The model based on Logistic Regression with an accuracy of 86% is chosen since it is ensured that it is not memorizing.
Description
Keywords
Machine learning, Balanceo, Energía eléctrica, Instalaciones eléctricas, Automatización, Mejoramiento continuo, Calidad en el servicio, Datos atípicos, Regresión Logística, Sobreajuste, Sistema predictivo, PCA (Análisis de Componentes Principales), Automation, Predictive system