Cross-validated tree-based models for multi-target learning.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Multi-target learning (MTL) predicts multiple outcomes simultaneously, using diverse methods from linear models to deep learning.
• Existing MTL techniques often struggle to effectively utilize correlations between targets, potentially limiting prediction accuracy and interpretability.

Purpose of the Study:

• To introduce a novel, interpretable, tree-based multi-target learning (MTL) scheme.
• To exploit correlations between multiple targets for improved prediction accuracy.
• To provide a method that avoids overfitting by using cross-validated splitting criteria.

Main Methods:

• A novel tree-based multi-target learning (MTL) approach is proposed.
• Cross-validated splitting criteria are employed at each tree node to identify correlated targets.
• The scheme integrates target correlations directly into the tree-building process.

Main Results:

• The proposed tree-based MTL scheme demonstrates significant performance improvements over alternative methods.
• Experiments on synthetic and real-world datasets validate the effectiveness of the approach.
• The method successfully leverages target correlations to enhance prediction accuracy while mitigating overfitting.

Conclusions:

• The novel tree-based MTL scheme offers a highly interpretable and accurate method for simultaneous prediction of multiple targets.
• Exploiting target correlations through cross-validated splitting is an effective strategy for improving MTL performance.
• The publicly available implementation facilitates further research and application of this advanced MTL technique.