SCIENCE@home

Causal Automated Machine Learning for Zero-Shot Decision-Making in Low-Resource Environments: A New Paradigm in Machine Learning Automation and Transferability

A. Patankar,P. Patil,Mihir Brahmane,Aaditya Meher,Anuj Maheshwari

Published 2025 in Cureus Journal of Computer Science

ABSTRACT

Automated machine learning (AutoML) has reduced the burden of manual model development, yet existing systems remain predominantly black-box predictors lacking causal reasoning. These limitations become particularly problematic in resource-constrained settings such as rural healthcare facilities, minority language processing, and emergency response operations. To address these challenges, we introduce a novel causal AutoML framework that integrates zero-shot learning for robust decision-making under minimal supervision. Our approach fundamentally shifts from prediction-focused black boxes to interpretable, causal-aware systems. We achieve this by embedding structural causal models directly into the AutoML pipeline, ensuring that model selection and optimization are guided by causal principles, not just statistical correlations. A key innovation is our causal-aware transfer engine, which uses graph-based contrastive learning to identify and transfer deep causal relationships rather than superficial feature similarities. This overcomes a critical failure point in traditional domain adaptation methods. We tested the framework using established benchmarks: Medical Information Mart for Intensive Care-III for healthcare, Infant Health and Development Program (IHDP) for policy evaluation, and Task-Aware Representation of Sentences (TARS) for natural language tasks. When tested on established benchmarks, our framework demonstrated significant performance gains in data-scarce conditions. On the IHDP policy evaluation benchmark, it achieved a precision in estimation of heterogeneous effects score of 1.6, and on the zero-shot TARS natural language processing benchmark, it outperformed Generative Pre-trained Transformer 3 by 2.7%. These statistically significant improvements (p < 0.05) highlight a practical path toward developing more reliable, interpretable, and ethically aligned AI systems for high-stakes applications where data is scarce and transparency is paramount.

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-31 of 31 references · Page 1 of 1

CITED BY

  • No citing papers are available for this paper.

Showing 0-0 of 0 citing papers · Page 1 of 1