This research introduces a novel mask-based output layer for multi-level hierarchical classification that enforces consistency and fairness across class hierarchies.

• Eliminates taxonomic inconsistencies by respecting hierarchical relationships between classes
• Enables flexible adaptation to new tasks without changing backbone architecture
• Incorporates fairness protections for sensitive attributes like age, gender and race
• Demonstrates superior predictive reliability in education, healthcare, and security domains

For educational applications, this approach ensures more reliable student classification and prediction systems while maintaining fairness across demographic groups—critical for equitable educational assessment and intervention planning.

Enforcing Consistency and Fairness in Multi-level Hierarchical Classification with a Mask-based Output Layer