Multiclass Oblique Random Forests With Dual-Incremental Learning Capacity.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Oblique Random Forests (ObRFs) are gaining traction for their efficiency in learning oblique hyperplanes.
• Existing ObRF methods primarily use off-line, batch training, lacking adaptability to new data or classes.
• Efficient dual-incremental learning (DIL) strategies for ObRFs remain underexplored.

Purpose of the Study:

• To develop an Oblique Random Forest with dual-incremental learning (DIL) capacity for on-the-fly classification.
• To enable the model to adapt to new input samples and new classes without complete retraining.
• To enhance the practical applicability of ObRFs in dynamic data environments.

Main Methods:

• A batch multiclass ObRF (ObRF-BM) algorithm was proposed, utilizing a broad learning system and a multi-to-binary approach.
• An optimal oblique hyperplane is identified in a higher dimensional space, creating supervised clusters at each node.
• A DIL strategy (ObRF-DIL) was developed to analytically update model parameters for incremental data and class additions.

Main Results:

• The proposed ObRF-DIL demonstrates effective on-the-fly classification capabilities.
• The DIL strategy allows for efficient model updates without the need for laborious retraining from scratch.
• Experimental results on public datasets show the superiority of the proposed approach over existing methods.

Conclusions:

• The developed ObRF-DIL provides an efficient solution for on-the-fly classification with incremental learning capabilities.
• The method effectively handles both new data instances and the emergence of new classes.
• This work significantly advances the adaptability and practical utility of Oblique Random Forests.