Causally-Guided Automated Feature Engineering with Multi-Agent Reinforcement Learning
arXiv:2602.16435v1 Announce Type: new Abstract: Automated feature engineering (AFE) enables AI systems to autonomously construct high-utility representations from raw tabular data. However, existing AFE methods rely on statistical heuristics, yielding brittle features that fail under distribution shift. We introduce CAFE, a framework that reformulates AFE as a causally-guided sequential decision process, bridging causal discovery with reinforcement learning-driven feature construction. Phase I learns a sparse directed acyclic graph over features and the target to obtain soft causal priors, grouping features as direct, indirect, or other based on their causal influence with respect to the target. Phase II uses a cascading multi-agent deep Q-learning architecture to select causal groups and transformation operators, with hierarchical reward shaping and causal group-level exploration strategies that favor causally plausible transformations while controlling feature complexity. Across 15
arXiv:2602.16435v1 Announce Type: new Abstract: Automated feature engineering (AFE) enables AI systems to autonomously construct high-utility representations from raw tabular data. However, existing AFE methods rely on statistical heuristics, yielding brittle features that fail under distribution shift. We introduce CAFE, a framework that reformulates AFE as a causally-guided sequential decision process, bridging causal discovery with reinforcement learning-driven feature construction. Phase I learns a sparse directed acyclic graph over features and the target to obtain soft causal priors, grouping features as direct, indirect, or other based on their causal influence with respect to the target. Phase II uses a cascading multi-agent deep Q-learning architecture to select causal groups and transformation operators, with hierarchical reward shaping and causal group-level exploration strategies that favor causally plausible transformations while controlling feature complexity. Across 15 public benchmarks (classification with macro-F1; regression with inverse relative absolute error), CAFE achieves up to 7% improvement over strong AFE baselines, reduces episodes-to-convergence, and delivers competitive time-to-target. Under controlled covariate shifts, CAFE reduces performance drop by ~4x relative to a non-causal multi-agent baseline, and produces more compact feature sets with more stable post-hoc attributions. These findings underscore that causal structure, used as a soft inductive prior rather than a rigid constraint, can substantially improve the robustness and efficiency of automated feature engineering.
Executive Summary
This article introduces CAFE, a causally-guided automated feature engineering framework that leverages multi-agent reinforcement learning to construct high-utility representations from raw tabular data. CAFE achieves up to 7% improvement over strong AFE baselines and demonstrates robustness under distribution shift. By incorporating causal discovery and hierarchical reward shaping, CAFE produces more compact feature sets with stable post-hoc attributions. The findings suggest that causal structure can be used as a soft inductive prior to improve the efficiency and robustness of automated feature engineering. The proposed framework has the potential to enable AI systems to adapt to changing data distributions, making it a valuable contribution to the field of automated feature engineering.
Key Points
- ▸ CAFE reformulates AFE as a causally-guided sequential decision process
- ▸ CAFE uses a cascading multi-agent deep Q-learning architecture
- ▸ CAFE achieves up to 7% improvement over strong AFE baselines
Merits
Improved Robustness
CAFE demonstrates robustness under distribution shift, reducing performance drop by ~4x relative to a non-causal multi-agent baseline
Compact Feature Sets
CAFE produces more compact feature sets with stable post-hoc attributions
Efficient Feature Construction
CAFE achieves faster convergence and competitive time-to-target
Demerits
Complexity
The proposed framework requires a deep understanding of causal discovery, multi-agent reinforcement learning, and hierarchical reward shaping, which may be challenging for non-experts
Scalability
The framework may not be scalable to very large datasets, requiring further research on efficient implementation and distributed computing
Expert Commentary
The proposed framework is a significant contribution to the field of automated feature engineering, demonstrating the potential of causal discovery and multi-agent reinforcement learning to improve the efficiency and robustness of AFE. However, the framework's complexity and scalability limitations need to be addressed in future research. Additionally, the study's focus on tabular data may limit its generalizability to other data types. Nevertheless, the findings of this study have far-reaching implications for the development of AI systems and highlight the importance of incorporating causal structure in AFE.
Recommendations
- ✓ Future research should focus on implementing CAFE in real-world applications and exploring its scalability to very large datasets
- ✓ The development of AFE frameworks that incorporate causal structure and multi-agent reinforcement learning should be encouraged
