Comparação entre análise discriminante e regressão logística para dados financeiros

Abstract

In financial sectors, the statistical techniques Logistic Regression and Discriminant Analysis are efficient for making decisions, both for classifying and discriminating groups, and adjusting a model that best fits the data. In this sense, this work proposed to compare the performance of Logistic Regression and Discriminant Analysis models in classifying groups in a financial sector database. Data were collected from the University of California at Irvine repository, comprising 1,000 observations with 20 variables, including a response variable classified by good or bad payer. We used the R software and the Tidyverse and MASS packages to manipulate and analyze the data. The result shows that Logistic Regression had a higher predictive power of 72.02% compared to Discriminant Analysis of 70.90%, making it the preferred choice for modeling default risk. These results provide valuable insights for credit decision-making in financial institutions.