Skip to content

⚖️ Cost-sensitive Classification

Overview

Not all mistakes are equal!
In FungiTastic, confusing a poisonous species for an edible one can have severe consequences, whereas the reverse is less dangerous.
This benchmark evaluates models under custom loss/cost functions that reflect real-world priorities.


Use Cases

  • Safety-critical systems (e.g., mobile ID for foragers)
  • Research into cost-sensitive, risk-aware, or calibrated classifiers

Data & Splits

  • Uses standard FungiTastic splits (temporal)
  • Poisonous/edible labels and additional risk metadata provided

Evaluation Protocol

  • Metrics: Weighted error rates, custom cost matrices (see Appendix in the paper)
  • You may define your own costs for specific errors

Baselines & Results


Quick Start

  • Use provided cost matrix and baseline scripts
  • Extend to new tasks with your own risk functions