Uncertainty-Aware Authentication Model for Fog Computing in IoT

Abstract

Since the term "Fog Computing" has been coined by Cisco Systems in 2012, security and privacy issues of this promising paradigm are still open challenges. Among various security challenges, Access Control is a crucial concern for all cloud computing-like systems (e.g. Fog computing, Mobile edge computing) in the IoT era. Therefore, assigning the precise level of access in such an inherently scalable, heterogeneous and dynamic environment is not easy to perform. This work defines the uncertainty challenge for authentication phase of the access control in fog computing because on one hand fog has a number of characteristics that amplify uncertainty in authentication and on the other hand applying traditional access control models does not result in a flexible and resilient solution. Therefore, we have proposed a novel prediction model based on the extension of Attribute Based Access Control (ABAC) model. Our data-driven model is able to handle uncertainty in authentication. It is also able to consider the mobility of mobile edge devices in order to handle authentication. In doing so, we have built our model using and comparing four supervised classification algorithms namely as Decision Tree, Naïve Bayes, Logistic Regression and Support Vector Machine. Our model can achieve authentication performance with 88.14% accuracy using Logistic Regression.

Risk-Based Adaptive User Authentication for Mobile Passenger ID Devices for Land/Sea Border Control

“…Towards this direction, effort should be placed on developing and implementing novel and efficient quantitative security risk estimation algorithms, suitable for sensitive applications. In the literature, various classification algorithms, such as decision trees [26]- [29], Naïve Bayes [16], [27], logistic regression [24], [27], etc. and other approaches, such as fuzzy logic [26], [30], [31] and Monte Carlo simulation [26], [32] have been proposed for quantitative risk estimation.…”

Risk-Based User Authentication for Mobile Passenger ID Devices for Land and Sea Border Control

“…Decision tree is a supervised classification algorithm that performs hierarchical decision making on the feature values based on a set of rules presented in a tree-like structure [26], [27]. As in all machine learning algorithms, firstly, data are divided into training and validation data sets.…”

“…In particular, any decision made divides the tree based on a criterion in a way that the training data is split into two or more branches. The main objective is to find the optimal split criterion so as the number of mixing the class variables in each branch of the tree is reduced as much as possible [27]. Afterwards, validation data are used to validate the decision tree and make necessary adjustments to the tree in order to make it more efficient [26], [28].…”

“…Reference [27] reports that there are three classical algorithms for implementing a decision tree: ID3, C4.5 and CART (Classification and Regression Trees). These algorithms employ 'Entropy' and 'Gini' as splitting criteria [27].…”

“…Reference [27] reports that there are three classical algorithms for implementing a decision tree: ID3, C4.5 and CART (Classification and Regression Trees). These algorithms employ 'Entropy' and 'Gini' as splitting criteria [27]. However, since the desired output of their system is the binary decision: Accept or Deny authentication, the authors in [27], among the three aforementioned classical algorithms for decision tree, applied the CART classification algorithm in order to build their data model.…”