| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.77 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O churn, ou seja, a saída de um cliente da empresa para outra concorrente, é uma
temática que impacta diversas indústrias/ negócios, não sendo a Galp uma exceção. Para
as empresas é instrumental detetar, antecipadamente, esse acontecimento, permitindo
assim a criação de iniciativas de retenção dos clientes propensos a realizar churn. Este
trabalho tem como objetivo criar um modelo de churn eficiente para as contas de gás
natural e eletricidade da Galp, utilizando dados dos contratos dos seus clientes. Após a
extração dos dados, procede-se ao seu pré-processamento e à sua divisão em três
conjuntos: Power (para as contas que apenas contratam o produto de eletricidade), Gas
(para as contas que apenas contratam o produto de gás natural) e Gas&Power (para as
contas que contratam ambos os produtos). Para a análise do churn, foi empregue o modelo
de logit leaf model, conhecido por LLM, que combina dois algoritmos - regressão
logística (RL) e árvores de decisão (AD). Além disso, foi testado como as partes
individuais do LLM, isto é, a RL e as AD, se comportam perante os conjuntos de dados,
utilizando as componentes da Análise de Componentes Principais (ACP) como forma de
eliminar a multicolinearidade nos dados. Os três modelos demostram sucesso em prever
o churn, tendo o LLM, em geral, obtido melhores resultados do que cada uma das suas
partes isoladamente.
Churn, that is, the departure of a customer from the company to a competitor, is an issue that impacts several industries/businesses, with Galp being no exception. For companies, it is crucial to detect and anticipate this event in order to create retention initiatives for customers prone to churn. The goal of this work is to create an efficient churn model for Galp's natural gas and electricity accounts, using data from customer contracts. After extracting the data, it was pre-processed and divided into three sets: Power (for accounts that only contract the electricity product), Gas (for accounts that only contract the natural gas product), and Gas&Power (for accounts that contract both products). For the churn analysis, the logit leaf model, known as LLM, was used, which combines two algorithms - logistic regression (LR) and decision trees (DT). Furthermore, it was tested how the individual parts of the LLM, i.e., LR and DT, behave in relation to the data sets, using the Principal Component Analysis (PCA) components to eliminate multicollinearity in the data. All three models showed success in predicting churn, with LLM, in general, achieving better results than each of its isolated parts.
Churn, that is, the departure of a customer from the company to a competitor, is an issue that impacts several industries/businesses, with Galp being no exception. For companies, it is crucial to detect and anticipate this event in order to create retention initiatives for customers prone to churn. The goal of this work is to create an efficient churn model for Galp's natural gas and electricity accounts, using data from customer contracts. After extracting the data, it was pre-processed and divided into three sets: Power (for accounts that only contract the electricity product), Gas (for accounts that only contract the natural gas product), and Gas&Power (for accounts that contract both products). For the churn analysis, the logit leaf model, known as LLM, was used, which combines two algorithms - logistic regression (LR) and decision trees (DT). Furthermore, it was tested how the individual parts of the LLM, i.e., LR and DT, behave in relation to the data sets, using the Principal Component Analysis (PCA) components to eliminate multicollinearity in the data. All three models showed success in predicting churn, with LLM, in general, achieving better results than each of its isolated parts.
Description
Keywords
Churn LLM Multicolinearidade ACP Multicollinearity, PCA
Pedagogical Context
Citation
Almeida, Inês de Sousa (2024). “Construção de um modelo de Churn dos clientes de gás e power da Galp” na intenção de compra e na atitude em relação à marca do consumidor”. Dissertação de Mestrado. Universidade de Lisboa. Instituto Superior de Economia e Gestão
Publisher
Instituto Superior de Economia e Gestão