Artificial Intelligence-Based Classification of Trusted and Untrusted Sensor Nodes in WBAN Using Multi Layered Stacked Naïve Bayes Method for Resilient Infrastructure

Human disease monitoring, identification, forecasting, and diagnosis have been made possible by wireless body area networks (WBANs). The potential applications of WBAN in the healthcare industry are described and defined throughout the paper. Privacy and security can be enhanced by WBANs, but their use is a concern. Using Artificial Intelligence-Based Classification on Sensor Nodes in WBAN, we propose a Multi-Layer Stacked Nave Bayes Method for Resilient Infrastructure to help overcome these issues. Patients can continue their daily routines while receiving medical care thanks to WBANs, which reduce foundation costs and pharmaceutical costs [1]. There is however a possibility that WBANs can fail to diagnose and treat patients accurately due to untrusted sensor nodes. WBAN networks can be made more efficient by identifying untrusted sensors. In this method, MLSNB is used to extract and categorize trust attributes from sensor nodes. MLSNB is the method that was used in this study and is described and explained in detail in this paper. Comparing MLSNB with other current methods, it also shows superior performance [2]. A min-max normalization technique was used to prepare the data. By using independent component analysis, we were able to extract features. For feature selection, we used a bi-objective genetic algorithm. To categories the nodes, we utilized Ensemble Naive Bayes. The paper improves healthcare services by identifying untrusted sensor nodes, the accuracy and reliability of the data collected by WBANs can be improved, which can lead to better diagnoses and treatments. This can ultimately lead to more efficient and effective healthcare systems that provide top-notch, affordably priced, and easily accessible health care services to the world's rising population, particularly those who are older and afflicted by ailments of old age [3]. In a multifaceted, dynamic, and connected setting like the Smart Home, where personal information is now more readily accessible from a distance, the renter is subject to privacy and security issues. A health care system can be installed in a smart home using wireless communication technologies, wearable sensor nodes, actuator nodes, and actuator nodes. A WBAN is the name of this technique. The architecture of WBAN is shown in Figure 1.

IoT improvements have made data detection, collection, and analysis processes more effective and efficient. IoMT, or "Internet of medical things," is a term that describes the usage of IoT and wireless body area networks in the healthcare industry. A variety of medical devices and software programmes with the ability to link to healthcare IT systems are referred to as IoMT [4]. WBAN evolves as a result of extensive research. A person's (outdoor) physical fitness, blood pressure, and risk of having a heart attack can all be analyzed by utilizing a sensor, cloud, and machine learning approach, as an example. Installing sensors (retina artificial arm chips) could give the blind individual their vision back. Research on heart problems, pneumonia, hypertension, Alzheimer's, Parkinson's, and other conditions can be done with WBAN. Although patients must stay in hospitals as part of traditional wellness programs, WBAN admits that these patients can continue with their regular daily routines. It reduces foundation costs and costs associated with pharmaceutical work [5]. Untrusted nodes might have a negative effect on the performance of a WBAN network as a whole. Because of outside attackers, the WBAN network topology becomes one of untrusted sensor nodes. This paper makes the following contributions:

·To prepare the data, we utilized min-max normalization technique.

·To extract features, we utilized independent component analysis.

·For feature selection, we used a bi-objective genetic algorithm.

·To categories the nodes, we utilized Ensemble Naive Bayes.

image003.jpg

Figure 1. Structure of WBAN

The remaining parts of the paper are provided below:

The related works are included in Section II. Section III provides specifics on the suggested technique along with the experimental data sets. A discussion of the result and analysis is provided in Section IV. The conclusion is presented in Section V. Healthcare is one of the industries that could benefit from this study. A wireless body area network (WBAN) holds considerable promise for monitoring, identifying, forecasting, and diagnosing human disease. Researchers found that WBAN networks can be improved to provide better healthcare through the use of the study's findings. The data collected by WBANs can be improved with better accuracy and reliability by identifying untrustworthy sensor nodes, which could lead to improved treatments and diagnoses. Throughout the introductory section of this paper, the objectives, methodology, and results of the study are outlined. WBAN networks can be improved by identifying untrusted sensor nodes as part of the study. Sensor nodes are classified into trusted and untrusted based on their trust aspects using the MLSNB method. Study results show that MLSNB performs better than other currently used methods. It is possible to use the study's findings to develop more efficient and effective healthcare systems that provide quality healthcare services at an affordable price to the world's growing population, especially those who are older and have chronic illnesses. Following is a breakdown of the remaining sections of the paper. This study presents a description of the experimental data sets and the proposed methodology in Section 2. In section 3, we present a methodology that we are proposing. Results and discussion are presented in Section 4. To conclude, Section 5 outlines directions for future research and provides a conclusion.

To categorize nodes within WBAN networks into trustworthy or untrusted, this study used an MLSNB algorithm. Sensor trust information is extracted from sensors using MLSNB by using an objective genetic algorithm (BGA) based on independent components extracted from each sensor. Using trust versus untrust classifications, WBAN networks can operate more efficiently. A performance record over other methods led to the selection of MLSNB. With the help of ICA and BGA, MLSNBs can categorize and extract relevant features from sensor nodes. The MLSNB implementation was evaluated against other current methods to provide more detail. The performance of MLSNB approaches makes them attractive in WBAN networks. This approach, however, may have limitations or challenges. It is possible that MLSNB's scalability may be compromised due to its higher computing resource requirement. As well as having broader implications and applications, this research has broad implications The WBAN network improves healthcare services for the elderly in a growing global population. By improving the performance of these networks, healthcare facilities can provide more comprehensive healthcare services. As an additional application, the MLSNB method may be useful in other fields beyond healthcare, such as environmental monitoring or industrial automation. This method can be used for both testing, which determines how well each node in the WBAN network functions, and training, which generates the input pattern. Figure 2 illustrates the WBAN process flow.

image004.png

Figure 2. Block diagram of proposed methodology

The WBAN network responses, or acquired signal levels, for typical human movements were gathered in order to train the classifier in the human motion categorization system. Based on earlier research, a WBAN transmission network was created in the bandwidths of 403.5 MHz and 2.45 GHz [16-22]. Data preprocessing modifies the data's format to speed up and improve the efficiency of processes like data mining and machine learning. The approaches are often used early in the AI and machine learning continued development and utilized to ensure high quality output [23-28]. Here, the Min-Max normalization step of the WBAN preprocessing procedure was applied. The unique records are converted linearly on this technique of records normalization. The minimal and most cost from the records is retrieved, and every cost is changed the usage of the Eq. (1):

 ${V}'=\frac{Min~A}{Max~A-MIN~A}(new\_\text{max }\!\!~\!\!\text{ }(A)-new\_min\left( A \right)+~new\_min\left( A \right)$                           (1)

3.1 Feature extraction using independent component analysis

A high order statistical unsupervised learning algorithm is called the ICA. The linear transformation is best estimated from the data itself since it may then be fine-tuned for the specifics of the input. In the case of regular transformations, the basis vectors do not adapt to new information but remain constant [29-38]. They can utilize a statistical latent variable model to define ICA. As random variables are combined linearly. s₁, ..., s_n that they observe, the model n random variables x₁, ..., x_n in (2):

${{x}_{i}}={{a}_{i}}{{s}_{1}}+{{a}_{2}}{{s}_{2}}+\ldots +{{a}_{in}}{{s}_{n}}$                                   (2)

when there exist some real coefficients, a_ij and i, j=1, ..., N. The S_i are statistically independent of one another by definition. The basic ICA model is presented here. It is not feasible to directly witness the independent components s_i since they are latent variables. Additionally, it is believed that the mixing coefficients a_ij are unknown. We only have the random variable x_i to go on, so we must use x_i to estimate the independent components s_i as well as the mixing coefficients a_ij. Under a general set of assumptions, this is possible. The mixing model is expressed as follows in vector-matrix notation (3):

$X=As=\underset{i=1}{\overset{n}{\mathop \sum }}\,{{a}_{i}}{{s}_{i}}$                          (3)

This model is viewed as a linear combination of basis vectors when it comes to image processing. Moreover, the basis vectors a_i is undoubtedly spatially, directionally, and spectrally localized. Finding a strong set of basis vectors is our aim to adequately depict iris patterns.

3.2 Feature selection using bi-objective genetic algorithm

To select the most beneficial and condensed feature subset, this research suggests using a bi-objective genetic algorithm (BGA) with a mutation pool. Figure 3 schematically depicts the suggested Bi-objective genetic algorithm (BGA)-based feature selection algorithm (FS). This suggested stochastic search-based feature selection technique is initially utilized to construct a subset of features from the dataset. Following the evaluation of these features, the non-dominated feature subset is selected.

image008.png

Figure 3. Diagram of one feature selector's flow

Due to the fact that GA is a population-based stochastic search technique, the initial population for the suggested feature selection is generated at random. P(n), where P is the expected population size and n is the total number of features in the dataset; together make up the dimension of the population set. The chromosomes were represented in this study as binary strings. The dataset's overall feature count is the same as the length of the string. For each gene, there are only two possible values, 1 and 0, which represent the presence or absence of the index feature in the current subset. Therefore, the following is a representation of chromosome c₁ of length K.

Every people have an equal chance of becoming a parent under the planned feature selection. To produce offspring from parent chromosomes, a single-point crossover technique is used. The application of mutation is mostly done to increase population variety. Multiple mutation strategies are used in the method, including bit flip mutation, leaping gene, Gaussian mutation, uniform and non-uniform mutation approach, and others. The mutation strategy that will be used at any given time for any current offspring is dynamically selected from the pool. By maximizing population member quality through the effective use of those individual mutual strategies, this mutation strategy was chosen. The suggested feature selection algorithm is listed below:

3.3 Multi layered stacked nave bayes (MLSNB)

This section illustrates the proposed Naive Bayes-based trust node classification method. The naive Bayesian model is straightforward along with the less complex model. The absence of repetitive parameter estimation shows that the naive Bayes model has simple functionality and is especially beneficial for very large data sets. One definition of the Bayesian Theorem is shown in (4) to (6):

P (x |y)=$\frac{\text{p}(\text{y}|\text{x})\text{ }\!\!~\!\!\text{ }\times \text{ }\!\!~\!\!\text{ p}\left( \text{x} \right)\text{ }\!\!~\!\!\text{ p}\left( \text{y} \right)}{\text{p}\left( \text{y} \right)}$                   (4)

$P~(y|x)=p~(y1~|x)\times p~(y2~|x)\times ....~\times ~p~({{y}_{n}}~|x$)                        (5)

where, P(x) is the probability of class occurrence, P(y) is the likelihood that instance y will occur, and P(y|x) is the likelihood that instance y will occur given class x given in the (4). All parameters are freed by the naive Bayes classifier. The Naive Bayes classifier can be used to simplify the Bayesian theorem in cases where there are multiple features or parameters. In this study, rogue WBAN nodes were found using a Naive Bayesian classifier. The Naive Bays classifier have been denoted by the “Packet Error Rate” (PER) and “Packet Loss Rate” (PLR). The hostile node intentionally discards packets or spreads false packets. It served as the impetus behind our choice to combine PLR and PER we suggested using the trust model to assess a sensor node's dependability. If PER=a and PLR=b for a node n under a piece of evidence. It served as the impetus behind our choice to provide PLR and PER in the trust model we suggested to assess a sensor node's dependability.

$Value=High:P(T=High\times P\left( PER=a=High \right) \times ~P~(PLR~=~b|T~=~High) \\ value=Low:~P\left( T=Low \right)\times ~P(PER=a|T=Low)\times (PLR=b|T=Low)  \\ Value=~Moderate:~P\left( T~=Moderate \right)~\times ~P(PER~=a|T=Moderate)   \times P\left( P=b\text{ }\!\!|\!\!\text{ }=Moderate \right)$                (6)

Trust worthiness is represented by the letter T. If the results of the calculation show that node n is malevolent, the expected value will be Low, followed by the predicted values and the forecast value. They used the Naive Bayes Classifier to classify the nodes' trustworthiness as high, moderate, and low. The trustee node with the highest trustworthiness value will subsequently be chosen by the coordination node. The coordination node will choose the trustee node with a moderate trustworthiness value if a high trustworthiness node is not present. Any trust node with a low trustworthiness rating will be permanently removed by the coordinator node and marked as hostile. In reaction to this, the coordinator node will then generate a notification.