Bayesian Neural Networks and K-Fold Causal BART for CATE Estimation

• Grant number: 24/06274-0
• Support Opportunities: Scholarships in Brazil - Master
• Start date: July 01, 2024
• End date: January 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Statistics
• Principal Investigator: Francisco Louzada Neto
• Grantee: Hugo Gobato Souto
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

This Master's research proposal focuses on pioneering the use of Bayesian Neural Networks (BNNs) for Conditional Average Treatment Effect (CATE) estimation and on improving the use of the Bayesian Additive Regression Tree (BART) model for CATE estimation, leveraging the framework proposed by Hahn et al. (2020). It addresses the notable gap in literature concerning BNN applications in CATE estimation by proposing a methodological advancement that combines the theoretical and practical benefits of BNNs over traditional models. Additionally, it improves the current BART-based models for CATE estimation by incorporating a K-fold approach to diminish over-fitting issues related to the BART model, creating the K-Fold Causal BART model. The research aims to validate the K-fold Causal BART and BNNs' efficacy in CATE estimation through rigorous empirical experimentation, comparative analysis against established models, and the development of user-friendly open-source tools to help the adoption of the proposed framework in Health and Social Sciences, especially by industry practitioners and researchers who do not possess a great amount of coding knowledge. The project underscores the potential of over-fitting strategies and BNNs to enhance scalability, speed, theoretical robustness, and generalization in CATE estimation, contributing significantly to the field of causal inference and offering new avenues for future research in machine learning and statistics.