iCVI-ARTMAP: Using Incremental Cluster Validity Indices and Adaptive Resonance Theory Reset Mechanism to Accelerate...

Area of Science:

• Computational Intelligence
• Machine Learning
• Data Mining

Background:

• Unsupervised learning models require robust methods for cluster assignment.
• Traditional clustering algorithms can be computationally intensive.
• Adaptive Resonance Theory (ART) offers a framework for incremental learning and pattern recognition.

Purpose of the Study:

• To present an enhanced Adaptive Resonance Theory Predictive Mapping (ARTMAP) model, termed iCVI-ARTMAP.
• To improve unsupervised learning by integrating incremental cluster validity indices (iCVIs) into the ART-based model.
• To reduce the computational burden of traditional offline clustering methods.

Main Methods:

• Developed the iCVI-ARTMAP model by incorporating iCVIs into the decision-making and mapping capabilities of an ART-based system.
• Implemented intelligent operations: sample assignment swapping, cluster splitting/merging, and variable caching for efficient iCVI recomputation.
• Utilized recursive formulations to minimize computational load.
• Created six iCVI-ARTMAP variants by integrating various iCVIs (one information-theoretic, five sum-of-squares-based) into fuzzy ARTMAP.

Main Results:

• The iCVI-ARTMAP model demonstrated improved incremental sample assignment to clusters.
• Experiments showed iCVI-ARTMAP outperformed or matched three ART-based and four non-ART-based clustering algorithms on benchmark datasets.
• Performance was dependent on the chosen iCVI and its suitability for the specific data.

Conclusions:

• iCVI-ARTMAP offers an efficient and effective approach to unsupervised learning and clustering.
• The model's general framework allows for easy integration of various iCVIs, offering flexibility.
• This adaptive resonance theory predictive mapping model presents a significant advancement in intelligent data analysis.