Discussion, Results and Analysis

In this study, the power load profile data is not grouped based on certain criteria, but it is grouped generally based on the pattern of power usage. Because the number of clusters to be used is unknown, the authors decided to use several clusters for the К-Means method grouping test. The cluster set used is from 2 cluster sets to 6 cluster sets. After calculating using the К-Means method [22], the optimization level of each cluster is then calculated using the DBI method. Following are the results of calculations on the last iteration as shown in Tables 11.5-11.12.

TABLE 11.5

The Number of Members of Each Iteration Set Cluster 6

TABLE 11.6

The End Centroid Value of Cluster 1 in Cluster Set 6

TABLE 11.7

The End Centroid Value of Cluster 2 in Cluster Set 6

TABLE 11.8

The End Centroid Value of Cluster 3 in Cluster Set 6

TABLE 11.9

The End Centroid Value of Cluster 4 in Cluster Set 6

TABLE 11.10

The End Centroid Value of Cluster 5 in Cluster Set 6

TABLE 11.11

The End Centroid Value of Cluster 6 in Cluster Set 6

TABLE 11.12

The Results of the DBI Value Calculation for Each Set of Clusters

Based on the calculation results in the application, the cluster set is the most optimal is set cluster 4 because it has the smallest DBI value, that is, 0.893, that means set cluster 4 has the density of each object with the best centroid and the distance between the clusters is also well separated [23].

After getting the optimal set of clusters, next is the testing phase. At this step, the data being tested is data of 3 customers categorized as customers with non-normal usage of electricity power. The test is, by determining the distance of each data testing object to each centroid in the cluster 4 set then the 3 data are tested in the application with the output that is, the 3rd data is not normal in electricity power usage. The test is, by determining the distance of each data testing object to each centroid in the cluster 4 set then the 3 data are tested in the application with the output that is, all 3 data are classified into customers with unnatural usage because based on the data allocation process to the centroid set the closest cluster 4, the distance of the testing data exceeds the maximum distance of each object in each cluster in the cluster 3 set [24,25].

General Detection Issue

As a general need for defense machinery in the wireless ad-hoc network to be integrated with IoT, several issues arise both related to the accuracy of defense machinery, the possibility of implementation, and others. Some issues related to this include [26]:

• • Accuracy: defense mechanism can detect Sybil at each phase with different properties. It must be able to discover large percentage of Sybil nodes to eliminate damage.
• • Cooperative Sybil detection: to detect effectively, all nodes in networks participate independently in the Sybil node detection process [27].
• • Low overhead costs: the proposed approach works more efficiently and requires fewer system resources.
• • Does not need for additional hardware at high prices.
• • Does not increase message exchange on the network.
• • Does not require much memory.
• • Detection time: the time needed to find and delete a Sybil entity is an essential factor that must be minimal.
• • Implementation: every IoT implementation such as in industry, smart home, smart grid, and others, there are special needs that must be considered in applying defense machinery [28].

FIGURE 11.3 Graphs of power consumption customers are not normal.

VANET Issue

In the wireless ad-hoc network area, VANET has become the most talked about topic lately, with specific needs that VANETs require additional requirements for security guarantees. Issues discussed in several papers reviewed are [29]:

• • Privacy Issue: most vehicle users hope that their identity information can be stored in VANET because they are afraid that their trip will leak with that identity.
• • Safety Issue: VANET does not allow a decrease in reputation after a severe traffic accident to prevent another attack, because damage to life and things in this attack cannot be repaired.

Learning-Based Issue

Defense machinery in the IoT infrastructure must be prepared with the needs of a “smart” system so that the application of scientific fields on artificial intelligence, especially machine learning is very open. Machine learning that needs to be applied to Sybil's defense machinery include: [30]

• • Deep Learning: with the development and the number of entities in an IoT infrastructure, a mechanism based on thorough analysis is needed, and deep learning has been successfully used in various areas including intrusion detection systems.
• • Online Learning: most of the data sent on IoT infrastructure, including WANET-based IoT, is a data stream, so online learning needs to be a concern for solutions on detection that continuously enhance the capability of defense machinery.

Centralized vs Decentralized Issue

• • Centralized issue: some defense machinery uses centralized detection, which requires a trusted center. Several papers on VANET build trust relationships that are bestowed on RSU. Installation of such infrastructure nationally is challenging to achieve in the early stages of VANET. Even in the medium term, there may still be many places that are not covered by RSU [31].
• • Decentralized issue: on the mechanism that relies on each node as a detector, all must know the credibility of each node that shares information around it and ensure all messages received are trusted and correct. However, this mechanism can work well assuming that most nodes are trusted nodes [32,33].