Ordinal regression based on learning vector quantization.

Area of Science:

• Machine Learning
• Ordinal Regression Analysis

Background:

• Ordinal regression is crucial for predicting ordered categories.
• Existing methods like ordinal generalized matrix learning vector quantization have limitations.
• A need exists for more effective ordinal regression techniques.

Purpose of the Study:

• To propose a novel ordinal regression approach within the learning vector quantization (LVQ) framework.
• To enhance existing LVQ-based ordinal regression with a more natural cost function and automatic bandwidth adaptation.
• To improve the accuracy and interpretability of ordinal regression models.

Main Methods:

• Development of a new ordinal regression model extending ordinal generalized matrix learning vector quantization.
• Introduction of a refined cost function for intuitive parameter updates.
• Implementation of automatic adaptation for prototype weight bandwidth.
• Empirical evaluation on diverse datasets.

Main Results:

• The proposed method demonstrates superior out-of-sample performance compared to alternative ordinal regression approaches.
• The new cost function leads to more intuitive and stable parameter updates.
• Automatic bandwidth adaptation enhances model flexibility and performance.
• Consistent improvements observed across various datasets.

Conclusions:

• The novel LVQ-based ordinal regression approach offers significant advantages over existing methods.
• The enhanced cost function and adaptive bandwidth contribute to superior predictive performance.
• This work provides a promising new direction for ordinal regression modeling.