ML / AI · Jun 2026
Privacy-Preserving Career Prediction Benchmark
A privacy-preserving multi-output career-prediction benchmark built on synthetic data, paired with an entropy-based feature selection routine that controls disclosure while preserving downstream predictive utility. The benchmark now includes McNemar and Cochran-Q significance tests, Cohen's kappa agreement, and per-class diagnostics.
Multi-output
Task
Synthetic
Data
Entropy-based
Selection
Problem
Career-trajectory prediction often relies on sensitive attributes, and naive feature use risks both leakage and unfairness. A benchmark was needed that pairs a privacy-respecting data generation process with a principled feature selection method.
Approach
I constructed a synthetic generative process for career trajectories, defined a multi-output prediction task with consistent splits, and applied entropy-based feature selection to retain informative features while removing those carrying disproportionate disclosure risk.
Results
The benchmark provides a reusable reference for evaluating multi-output career models under explicit privacy constraints and demonstrates that entropy-based selection preserves predictive utility while reducing leakage.
Stack