Year: 2021
Type: article
Abstract: In this paper, a novel algorithm (IBC1) for graph clustering with no prior assumption of the number of clusters is introduced. Furthermore, an additional algorithm (IBC2) for graph clustering when the number of clusters is given beforehand is presented. Additionally, a new measure of evaluation of clustering results is given—the accuracy of formed clusters (T). For the purpose of clustering human activities, the procedure of forming string sequences are presented. String symbols are gained by modeling spatiotemporal signals obtained from inertial measurement units. String sequences provided a starting point for forming the complete weighted graph. Using this graph, the proposed algorithms, as well as other well-known clustering algorithms, are tested. The best results are obtained using novel IBC2 algorithm: T = 96.43%, Rand Index (RI) 0.966, precision rate (P) 0.918, recall rate (R) 0.929 and balanced F-measure (F) 0.923.
Source: Mathematics
DOI: 10.3390/math9101089
Authors: Nebojsa Budimirovic, Nebojsa Bacanin
Institutions: Academy of Applied Studies Šabac, Singidunum University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,728
OpenAlex ID: W3161985013