# Balancing misclassification errors in image-based inference using problem domain semantics and a nested cascade architecture

## TL;DR

This paper introduces a method to reduce costly misclassification errors in machine learning by using class label semantics and a layered neural network architecture.

## Contribution

A novel nested cascade architecture that leverages class label semantics to prioritize error reduction in deeper hierarchical classes.

## Key findings

- A nested cascade architecture reduces error severity by prioritizing deeper hierarchical classes.
- Using class label semantics improves performance in image and tabular data tasks.
- Early layers in the cascade focus on simpler, less severe classification problems.

## Abstract

Pattern recognition models, particularly neural networks, often focus on maximising classification accuracy. However, in practice, the types of errors made (misclassification between different classes) can have varying associated costs. Current methods overlook varying misclassification error types. Misclassification labels can either be available from expert knowledge or derived from semantics of textual descriptions of class labels. Exploiting such misclassification costs can have significant implications when deploying machine learning systems. Here, using five examples from image and tabular domains, we show how a deep neural architecture trained in a nested layer-wise fashion (cascade learning) in which early layers solve easier problems than later ones could exploit such hierarchical aspects of class labels. We employ a measure of performance called “severity” of errors and show how emphasis could be placed on classes that are deeper in the hierarchy, ignoring errors that arise between semantic neighbours.

The online version contains supplementary material available at 10.1007/s00521-025-11613-8.