Application of Neural Networks for Decision Making and Evaluation of Trust in Ad-hoc Networks

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21240%2F18%3A00326796" target="_blank" >RIV/68407700:21240/18:00326796 - isvavai.cz</a>

Result on the web

<a href="http://pesw.fit.cvut.cz/2018/PESW_2018_2.pdf" target="_blank" >http://pesw.fit.cvut.cz/2018/PESW_2018_2.pdf</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Application of Neural Networks for Decision Making and Evaluation of Trust in Ad-hoc Networks

Original language description

The lack of infrastructure and central management in ad-hoc networks is an advantage from the viewpoint of scalability and flexibility, but it poses security risks and requires close cooperation among nodes for the network to function as a whole. Trust-based approach looks promising for improving security and cooperation. By the trust problem, we understand the problem of measuring the confidence in the fact that individual node will cooperate - by properly delivering the data in transit, sourced or destined for other nodes. We model trust using the packet delivery ratio (PDR) metric. We have developed a method to apply neural networks (NNs) for solving the problem of trust. It demonstrates that NNs are capable of detection of untrusted nodes and estimation of the trust values. We have conducted a series of simulation experiments and measured the quality of our method. Our experiments show clearly that NNs can be effectively used for solving the problem of detection of untrusted nodes and trust value estimation. The results show in average 98% accuracy of the classification and 94% of the regression problem. An important contribution of our research is a verification of the hypothesis that synthetic generation of ad-hoc network traffic in a simulator is sufficient for training of a NN that is then capable to accurately estimate trust. Our NN-based method can be applied in a running ad-hoc network with a given topology. Training of the NNs can be done without collecting data from a running network, since the training data can be constructed artificially. In case of topology changes, new learning of NN can be performed quickly and effectively. No active measurements is needed. The contributions of this paper are (1) confirmation of applicability of NNs to trust management in ad-hoc networks; (2) construction of a method to detect untrusted nodes and to estimate the value of trust using NNs.

Czech name

—

Czech description

—

Type

O - Miscellaneous

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

S - Specificky vyzkum na vysokych skolach

Publication year

2018

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů