From Data to Decision: Feature Selection as A Compliance Mechanism for Lawful, Fair, and Transparent Income Classification in Civil and Commercial Decision-Making

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Income-based automated decision systems increasingly influence civil and commercial outcomes such as credit approval, employment screening, insurance pricing, and tenancy assessment. While machine learning models can enhance predictive accuracy, they also raise civil and commercial law concerns relating to discriminatory impact, transparency, reason-giving, and liability for negligent or unfair automated decision-making. This paper evaluates whether feature selection can serve as a governance mechanism that improves model robustness while strengthening legal defensibility through reduced dimensionality and clearer feature rationales. Using adult dataset, this paper considered a customized Minimum Redundancy-Maximum Relevance (mRMR), customized machine learning models and assessed using key metrics, alongside compliance oriented interpretations of feature reliance. Ensemble models like Stacking and bagging models achieve strong performance with reduce computational cost and enhance interpretability. However, the recurring selection of legally sensitive or proxy-sensitive attributes highlights the need for fairness auditing, transparency documentation and human oversight. The study concludes with a civil-commercial governance framework for deploying income prediction models responsibly.