A new clustering method based on multipartite networks.

Area of Science:

• Machine Learning
• Data Mining
• Network Science

Background:

• Clustering is a fundamental machine learning task for discovering hidden patterns in data without prior labels.
• Network-based clustering algorithms are popular, typically representing data as graphs with nodes as data instances and edges as similarity measures.
• Existing methods face challenges in effectively capturing complex relationships within high-dimensional datasets.

Purpose of the Study:

• To propose a novel network-based clustering approach utilizing a multipartite network structure.
• To demonstrate the effectiveness of the proposed method in identifying data similarities through network path analysis.
• To evaluate the performance of the novel clustering method on synthetic and real-world datasets, including a health and population dataset.

Main Methods:

• The proposed method constructs a multipartite network where layers represent data attributes and nodes represent data intervals.
• Clustering is achieved by analyzing the information derived from paths within this multipartite network.
• The approach was tested on synthetic datasets, real-world benchmarks, and a World Bank dataset for country grouping.

Main Results:

• The novel multipartite network clustering approach demonstrated comparable performance to state-of-the-art methods.
• The method showed superior performance on specific datasets, highlighting its effectiveness in certain scenarios.
• Application to World Bank data successfully grouped countries based on health, nutrition, and population indicators.

Conclusions:

• The proposed multipartite network approach offers a promising new direction for clustering algorithms.
• The method provides an intuitive way to construct clusters based on network path information.
• The approach is versatile, applicable to diverse datasets and real-world problems, and competitive with existing techniques.