Graph Based Multi-Layer K-Means++ (G-MLKM) for Sensory Pattern Analysis in Constrained Spaces

Area of Science:

• Machine Learning
• Data Science
• Graph Theory

Background:

• Traditional data-target association methods rely on statistical probabilities.
• Tracking targets in constrained spaces with limited sensor data presents significant challenges.
• Existing methods struggle with complex environments and minimal target information.

Purpose of the Study:

• To develop a novel unsupervised machine learning algorithm for data-target association.
• To address the challenge of tracking targets in constrained spaces with limited sensor data.
• To improve the accuracy of data-target association by employing graph theory and clustering.

Main Methods:

• Developed multi-layer k-means++ (MLKM) for local data-target association.
• Proposed a p-dual graph to represent interconnected constrained spaces.
• Generalized MLKM to graph based multi-layer k-means++ (G-MLKM) using graph theory.
• Integrated error correction mechanisms to enhance association accuracy.

Main Results:

• The G-MLKM algorithm effectively solves the data-target association problem.
• Achieved an average data-target association accuracy of 92.2% in simulations.
• Demonstrated superior performance compared to traditional statistical methods.

Conclusions:

• G-MLKM offers a robust and accurate solution for data-target association in challenging environments.
• The unsupervised clustering approach outperforms traditional probabilistic methods.
• The integration of graph theory and error correction significantly improves tracking accuracy.