Adaptive neural tree exploiting expert nodes to classify high-dimensional data.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• High-dimensional data classification faces challenges like the curse of dimensionality and overfitting.
• Traditional neural trees can result in excessively deep trees or decreased accuracy when shallower trees are used.

Purpose of the Study:

• To introduce a novel Neural Tree exploiting Expert Nodes (NTEN) for improved high-dimensional data classification.
• To address the limitations of existing neural tree methods in handling data complexity and overfitting.

Main Methods:

• NTEN utilizes a decision tree structure with expert nodes capable of multi-dimensional splitting.
• Expert nodes select optimal neural networks, evaluate overfitting, and cluster features to minimize redundancy.
• Metaheuristic optimization algorithms (GA, NSGA-II, PSO, ACO) are employed for feature clustering and network selection.

Main Results:

• NTEN achieves good classification results on 35 standard datasets.
• The method successfully reduces tree depth without compromising classification accuracy or increasing overfitting.
• Expert nodes dynamically adjust strategies based on data complexity and overfitting levels.

Conclusions:

• NTEN offers an effective solution for classifying high-dimensional data, overcoming limitations of existing methods.
• The proposed expert node mechanism enhances decision tree performance in complex, high-dimensional scenarios.
• NTEN demonstrates a balance between tree depth, accuracy, and overfitting control.