Multi-objective evolutionary optimization for constructing neural networks for virtual reality visual data mining:...

Area of Science:

• Computer Science
• Data Mining
• Artificial Intelligence

Background:

• Virtual reality (VR) offers immersive environments for data exploration.
• Visual data mining requires effective methods for representing high-dimensional data.
• Current methods often rely on single-objective optimization, limiting representational capabilities.

Purpose of the Study:

• To develop a new method for constructing virtual reality spaces for visual data mining.
• To integrate supervised classification and unsupervised structure preservation within VR space construction.
• To enhance data representation by utilizing multi-objective optimization.

Main Methods:

• Utilized multi-objective optimization with genetic algorithms on nonlinear discriminant (NDA) neural networks.
• Employed two neural network layers for simultaneous supervised classification and unsupervised similarity preservation.
• Applied gene expression programming to derive analytic representations of the generated VR spaces.

Main Results:

• Constructed a set of VR spaces offering simultaneous solutions for classification and structure preservation.
• Demonstrated a conceptual improvement over single-objective optimization techniques.
• Successfully applied the domain-independent approach to geophysical prospecting for cave detection.

Conclusions:

• The proposed method provides a superior approach to constructing VR spaces for visual data mining.
• Simultaneous optimization objectives lead to richer and more informative data representations.
• The technique shows promise for diverse applications, including geophysical exploration.