Sensor fusion based on Dempster-Shafer theory of evidence using a large scale group decision making approach

Koksalmis E., Kabak Ö.

INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS, vol.35, no.7, pp.1126-1162, 2020 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 35 Issue: 7
  • Publication Date: 2020
  • Doi Number: 10.1002/int.22237
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, PASCAL, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Compendex, Computer & Applied Sciences, INSPEC, Metadex, zbMATH, Civil Engineering Abstracts
  • Page Numbers: pp.1126-1162
  • Keywords: classification, Dempster-Shafer theory of evidence, large scale group decision making, sensor fusion, sensor weighting, TARGET CLASSIFICATION, REASONING APPROACH, BELIEF, ALGORITHM, COMBINATION, UNCERTAINTY, RELIABILITY, TBM
  • Istanbul Technical University Affiliated: Yes


In group decision making (GDM), the quality of the solution relies primarily on the quality and the expertize of decision makers. At that point, deriving the weights, which reflects their importance or perceived reliability of decision makers, presents as a new challenge. In addition to that, the uncertainty is also a common problem for GDM. These problems are also faced in the sensor fusion problem where information from multiple sources must be aggregated. Therefore, in this study, a large scale GDM approach for sensor fusion is proposed. Since the proposed method is a clustering-based method, it provides acceptable results in the sensor networks consisting of multiple sensors. It can work under uncertainty as a result of converting the raw data obtained from sensors to the basic probability assignments. It also considers the reliability of the sensors clusters by assigning three objective weights. In addition to these objective weights, the proposed method enables to assign subjective weights to integrate supervisors/intelligence analyst experiences and knowledge in the problem field. The applicability and the validity of the proposed method are checked through two real classification data sets: ionosphere and forest type mapping data set. Experiments show that the classification rate is increased significantly when the proposed method is applied to two data sets. Finally, effect of extension parameter, objective weights, reliability threshold, number of clusters and clustering method on the classification rate and the detection probability are examined, and future studies are provided in conclusion.