If a multi-class classification model consistently yields high accuracy but performs poorly on a specific underrepresented class, what underlying issue might this indicate about the loss function used?

Question

Seekh · Accepted Answer

A high overall accuracy with a weak underrepresented class suggests the loss function is treating all classes the same, so it focuses on the majority classes that dominate the error signal. In plain terms, the loss gives little weight to mistakes on the minority class, so the model learns to predict the common classes well and ignores the rare one. This imbalance in the loss surface causes the model to ignore the minority class during training. For example, if 90 % of the data is class A and 10 % is class B, a plain cross‑entropy loss will mainly penalize errors on class A, letting the model overlook class B. The fix is to use a weighted or focal loss that gives higher importance to the underrepresented class.

If a multi-class classification model consistently yields high accuracy but performs poorly on a specific underrepresented class, what underlying issue might this indicate about the loss function used?

Learning Path

Choose the Best Answer

Understanding the Answer

Key Concepts

Deep Dive: Multi-class Loss Functions

Definition

Topic Definition

Ready to Master More Topics?