Banca de QUALIFICAÇÃO: Felipe Gonçalves Pereira

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : Felipe Gonçalves Pereira
DATE: 16/10/2023
TIME: 19:00
LOCAL: On-line
TITLE:

Predicting Inflation in Brazil: An Integration of Shrinkage Methods for Feature Selection and an Agnostic Interpretation of Forecast Results.

KEY WORDS:

Inflation, Machine Learning, Forecasting, Model Interpretability, Shapley Value

PAGES: 40
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Matemática da Computação
SPECIALTY: Modelos Analíticos e de Simulação
SUMMARY:

We employ machine learning techniques to predict the IPCA. Initial results show the superiority of Ridge regression over LASSO models, possibly due to correlated predictors such as commodity prices and inflation indices. While current research validates the effectiveness of survey methods in understanding IPCA inflation patterns, future work
emphasis will shift to projections using non-linear models, particularly Random Forest and Gradient Boosting models. The most relevant variables from the winning economic model with the complete data set will be used to carry out the selection of variables to be used in non-linear models. To deepen our analysis, we will use an agnostic approach to interpretability of the differences made by nonlinear models, particularly the Shapley Value technique. This objectively ensures more reliable results for decision making.

COMMITTEE MEMBERS:
Presidente - ***.366.561-** - JOAO GABRIEL DE MORAES SOUZA - UnB
Interno - 3164711 - JOAO CARLOS FELIX SOUZA
Interno - ***.535.421-** - PENG YAOHAO - UnB
Externo à Instituição - MATHIAS SCHNEID TESSMANN