Machine Learning Based Trust Computational Model for IoT Services

Upul Jayasinghe, Gyu Myoung Lee, Tai‐Won Um, Qi Shi

IEEE Transactions on Sustainable Computing · 2018 · 248 citations · 39 references

DOIFull text

Open access

Concepts

TL;DR

The Internet of Things generates vast amounts of sensitive data, exposing threats from data management to discriminatory analytics, making trust essential yet difficult to establish due to diverse cyber‑physical factors, thus necessitating an intelligent trust computation model. This study proposes a quantifiable trust assessment model for IoT services. The model numerically computes individual trust attributes, applies a machine‑learning algorithm to classify and combine these features into a final trust value, and validates its effectiveness through simulation. Simulation results demonstrate that the proposed method outperforms existing aggregation approaches.

Abstract

The Internet of Things has facilitated access to a large volume of sensitive information on each participating object in an ecosystem. This imposes many threats ranging from the risks of data management to the potential discrimination enabled by data analytics over delicate information such as locations, interests, and activities. To address these issues, the concept of trust is introduced as an important role in supporting both humans and services to overcome the perception of uncertainty and risks before making any decisions. However, establishing trust in a cyber world is a challenging task due to the volume of diversified influential factors from cyber-physical-systems. Hence, it is essential to have an intelligent trust computation model that is capable of generating accurate and intuitive trust values for prospective actors. Therefore, in this paper, a quantifiable trust assessment model is proposed. Built on this model, individual trust attributes are then calculated numerically. Moreover, a novel algorithm based on machine learning principles is devised to classify the extracted trust features and combine them to produce a final trust value to be used for decision making. Finally, our model's effectiveness is verified through a simulation. The results show that our method has advantages over other aggregation methods.

References

39