HyConEx: Hypernetwork classifier with counterfactual explanations for tabular data

Built-in explanations reduce the "black box" problem: Unlike most AI systems that make predictions first and require separate tools to explain them, HyConEx combines prediction and explanation in a single model, making it easier to understand why decisions are made without adding extra steps.

Counterfactual explanations show what needs to change: The model generates specific examples showing which features would need to be modified and by how much to change the outcome, providing actionable insights (e.g., "if your credit score were 50 points higher, the loan decision would be approved").

Designed specifically for real-world business data: HyConEx is built for tabular data—the structured spreadsheet-like information used in banking, healthcare, and HR decisions—rather than images or text, making it directly applicable to high-stakes domains where explainability matters most.

Maintains competitive accuracy while explaining: The model achieves strong performance on standard accuracy metrics while simultaneously providing interpretable explanations, demonstrating that transparency doesn't require sacrificing predictive performance.

Enables fairness auditing and contestability: By showing exactly which features drive individual decisions and what would need to change to alter outcomes, this approach supports organizations in auditing for bias and helps individuals understand and potentially contest automated decisions affecting them.