Label dependency modeling in Multi-Label Naïve Bayes through input space expansion.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Multi-label learning assigns multiple labels to instances, unlike traditional single-label methods.
• Existing multi-label techniques often use a shared feature space, neglecting unique label semantics.
• There is a need for methods that capture label-specific characteristics and correlations.

Purpose of the Study:

• To propose an improved multi-label Naïve Bayes (iMLNB) algorithm.
• To effectively model label correlations within a multi-label learning framework.
• To enhance classification by integrating meta-information from the label space.

Main Methods:

• Expanded the input space of Naïve Bayes to include meta-information from the label space.
• Created a composite input domain with continuous and categorical variables.
• Refined likelihood parameters using a joint density function to handle heterogeneous data types.

Main Results:

• The enhanced iMLNB model demonstrated superior performance compared to the traditional multi-label Naïve Bayes (MLNB).
• Empirical evaluation on six benchmark datasets confirmed the competitive edge of iMLNB.
• The proposed method showed significant improvements across various evaluation metrics.

Conclusions:

• Modeling label dependencies is crucial for effective multi-label learning.
• The iMLNB approach offers a robust and effective solution for multi-label classification.
• This work represents a significant contribution to the field of multi-label learning.