UntitledFederated Privacy-Preserving Feature Selection for University Moral Education Data Using Adaptive Spiral-Flight Sparrow Search Item

In multi-institutional settings, however, these records remain distributed across independently governed information systems, while differences in student populations and feature distributions complicate collaborative analysis. This study formulates feature selection in this setting as a horizontal cross-silo problem and proposes an adaptive spiral-flight sparrow search algorithm (ASFL-SSA) within a privacy-preserving federated framework.

The framework separates relevance guidance from candidate evaluation. Stage A constructs a global search prior from locally estimated mutual-information scores after clipping and Gaussian perturbation, whereas Stage B evaluates candidate feature masks using securely aggregated client-level validation AUC scores. ASFL-SSA combines prior-guided Tent-chaos initialization, adaptive producer allocation and inertia control, stagnation-triggered spiral exploration, opposition-based search, and budget-aware single-bit refinement.