Published Paper Details:

Abstract: Graph pattern mining is a vital research field. It includes the identification of interaction flows between compounds found in chemicals or genes. This is attained by representing these compounds as graph interactions, allowing researchers to evaluate and recognize complex relationships and patterns. Graph-based representations provide researchers with a powerful tool to uncover valuable insights and make informed decisions in various fields. Extracting the interaction flow from these graphs is a very challenging task. Data mining technique such as Clustering is used to groups graph nodes with significant communication to discover the graphical interaction flow. To address this, we propose the Irrelevant Feature-aware NMF Clustering Method (IF-NMFCM) for clustering graph nodes. This technique focuses on eliminating irrelevant features to improve clustering accuracy while reducing associated dataset complexities. It incorporates principal component analysis for data pre-processing and utilizes the Particle Swarm Optimization (PSO) method for feature selection, ensuring the exclusion of irrelevant features. By adopting this technique, we obtain an optimal feature set that enhances clustering accuracy. Finally, pre-processed data is represented as a graph and then clustered using the hierarchical NMF clustering method. Our proposed research work establishes superior results in terms of improved accuracy rate than the existing approaches. The research work is implemented in MATLAB simulation environment for comprehensive analysis.