• Computer Science > Artificial Intelligence [Submitted on 18 Feb 2026] Title:Causally-Guided Automated Feature Engineering with Multi-Agent Reinforcement Learning View PDF HTML (experimental)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.
Article Summaries:
- Summary
A new framework, CAFE, rethinks automated feature engineering (AFE) by embedding causal reasoning into a multi‑agent reinforcement learning pipeline. Phase I learns a sparse directed acyclic graph to assign soft causal priors to features relative to the target, categorising them as direct, indirect, or unrelated. Phase II employs a cascading deep Q‑learning architecture that selects causal groups and transformation operators, guided by hierarchical reward shaping and causal‑group exploration to favor plausible transformations while limiting feature complexity. Across 15 public benchmarks, CAFE outperforms strong AFE baselines by up to 7 % in classification and regression metrics, converges faster, and under covariate shift shows a four‑fold reduction in performance drop. The approach demonstrates that causal structure, used as a soft inductive prior, can markedly improve robustness and efficiency in automated feature construction.
Sources: