Banca de QUALIFICAÇÃO: GUSTAVO HENRIQUE FARIAS BEZERRA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : GUSTAVO HENRIQUE FARIAS BEZERRA
DATE: 18/03/2024
TIME: 08:00
LOCAL: Remoto
TITLE:

Predicting and Interpreting Churn: Integrating Causal Analysis and Machine Learning for Effective Retention Strategies

KEY WORDS:

Churn, CRM (Customer Relationship Management), Predictive Analysis, Machine Learning, Causal Inference.

PAGES: 62
BIG AREA: Engenharias
AREA: Engenharia de Produção
SUMMARY:

The interconnectedness of economies, widespread internet use, and the growth of economic globalization have reshaped the dynamics between consumers and industries, creating an active interaction between the two parties. In this context, understanding the customer lifecycle becomes essential for the operational and financial health of companies, focusing on analyzing the elements that influence customer satisfaction and loyalty. The phenomenon of churn, indicating customer loss, emerges as a significant challenge for various industries, directly affecting profitability and business sustainability. Therefore, the aim of this research is to perform churn prediction modeling, seeking not only greater accuracy but also interpretability, through the addition of causal analysis concepts. The Telco Customer Churn database, version 11.1.3 from IBM, was used as empirical support for the propositions. This study aims to identify factors that may influence customer loss and examine different retention strategies through interventions. The suggested methodology combines machine learning algorithms, such as gradient boosting (LGBM) and decision trees, with sophisticated causal analysis techniques, including Double Robust Machine Learning and conditional treatment effect modeling (CATE). The focus is not just on predicting churn using the best possible metrics but also on understanding the factors leading customers to end their relationship with the company, considering variables from profile characteristics to contracted services. Variables such as contract type, gender, age, dependents, contract duration, total monetary amount of bills paid, and discounts were analyzed. The expected results of this study aim to validate the hypotheses of Wu et al. (2012), especially regarding the prediction effect, and offer new insights into the profile of customers prone to leave the company. The conclusions of this work promise to significantly contribute to the field of customer relationship management, providing essential strategic data for the effective development of retention strategies to assist managers' decision-making.

COMMITTEE MEMBERS:
Interno - 1142787 - JOSE ALFREDO FERREIRA COSTA
Externo à Instituição - MARCUS VINICIUS DANTAS DE ASSUNCAO
Interna - 1777131 - MARIANA RODRIGUES DE ALMEIDA