Adaptive metric learning vector quantization for ordinal classification.

Area of Science:

• Machine Learning
• Pattern Analysis
• Data Mining

Background:

• Traditional classification methods often ignore inherent order in data classes, potentially reducing accuracy.
• Nominal classification schemes do not utilize ordinal relationships, leading to suboptimal performance in ordered classification tasks.

Purpose of the Study:

• To introduce two novel ordinal learning vector quantization (LVQ) schemes incorporating metric learning.
• To address the limitations of nominal classification by utilizing class order information in the learning process.

Main Methods:

• Developed two ordinal LVQ schemes that integrate metric learning.
• Modified the training process to utilize class order information for prototype selection and updates.
• Compared ordinal LVQ performance against nominal LVQ and benchmark ordinal regression models.

Main Results:

• The proposed ordinal LVQ formulations demonstrated favorable comparisons with their nominal counterparts.
• Experimental results showed competitive performance against existing ordinal regression models.
• Ordinal LVQ effectively utilizes class order information, enhancing classification accuracy.

Conclusions:

• The novel ordinal LVQ schemes provide an effective approach for classification problems with ordered classes.
• These methods offer a valuable alternative to nominal classification and existing ordinal regression techniques.
• Ordinal LVQ schemes enhance pattern analysis by incorporating crucial class order relationships.