Classtree works by recognizing that classes can be grouped into a hierarchy of superclasses and subclasses. This approach allows for more general predictions when specific details are uncertain, optimizing the balance between being correct and specific.

• Granularity Control for Precision: Efficient algorithm to balance specificity and correctness in classification, enabling more informed decisions at varying levels of detail without sacrificing accuracy
• Flexible and Accurate Modeling: Novel loss functions enables superior performance across diverse data structures, enhancing predictability.
• Robust to Emerging Categories: Demonstrating superior performance on unseen classes, maintaining high accuracy in classifications and predictions.

Classtree uses a threshold to vary the prediction's granularity, enabling a classifier to make confident guesses at a higher level in the hierarchy even if it's unsure about finer details. This strategy is particularly useful when exact leaf-node predictions are challenging, offering a flexible way to navigate through the class hierarchy based on confidence levels.