Secure heart disease prediction model using ESVO-based Swish Bessel CNN classifier

Abstract

Heart disease is a critical issue that affects people, causes serious sickness, and is the main cause of mortality worldwide. Early diagnosis of disease plays a significant role in heart disease prediction and is attained by various automation techniques. The availability of automation techniques initiates the necessity for medical data and the storage of medical data becomes a research problem due to its high sensitivity. The emergence of IoT networks formed a promising solution for data storage through the cloud server and preventing the data from various threats is a challenging problem. A secure heart disease prediction system is developed by the utility of the ESVO-based Swish Bessel CNN classifier (Emperor Spheniscidae Vampire Optimization-based Swish Bessel Convolutional Neural Network), and the important significance of the research depends on the ESVO optimization that helps in gaining a deeper insight of the classifier as well as helps in preventing the threatening of data. The security of the cloud server is enhanced by the EDH-ECC (Entropy Diffie Hellman – Elliptic Curve Cryptography) which promotes the information exchange even in unsecured channels. Similarly, the authentication and authorization of the cloud server are carried out using the EAN-13 and salt-based digital signature that initiates strong credentials and enhance data security. Finally, the heart disease is diagnosed using the ESVO-based Swish Bessel CNN classifier. Assessing the accuracy, sensitivity, specificity, and F1-measure, which provided values of 94.877 %, 95.464 %, 93.293 %, and 95.14 % shows the effectiveness of the research.

Keywords

Heart disease ESVO-based Swish Bessel CNN classifier EDH-ECC EAN-13 security

1. Introduction

Internet of Things (IoT) is an emerging technology that plays an indisputable role in the futuristic world and is formed by the invisible interconnection of the multiple objects that are connected to the environment. IoT is associated with numerous smart objects that are connected to the main device and effective computing is provided by these networks without manual intervention [22,26]. The applications of IoT cover a wide range of fields and the utility of IoT devices for the effective prediction of diseases in the medical field is in high demand. One of the diseases that threatened the life of multiple people is cardiovascular disease [11]. At the initial stage of old age, people are highly affected, and in recent times all age group people are affected. Cardiovascular diseases are the condition, where the malfunction of the heart or blood vessels occurs that causes severe chest pain, fatigue, and shortness of breath that causes discomfort ability to the patients [2,19]. Early diagnosis of this disease helps to decrease the mortality rate by taking countermeasures. The detection of cardiovascular disease with high accuracy initiates a necessity for a enourmous data and dwelling with this large data in the IoT devices are strenuous task due to the storage capability and sensitivity of the data [23,25]. Hence the storing of big data in the IoT devices is performed by the cloud architectures enabled in the IoT a device that ensures privacy by the utilization of various techniques [26].

The World Health Organization (WHO) conveyed that the highest death rate is due to coronary heart disease and around 31% of the population died due to this disease [26]. Although the diagnosis of coronary disease is a major issue, preserving the sensitive data of the patient to ensure its integrity is also an equivalently important task [12]. Enabling the cloud computing technology in the network permitted the users to decrease the computational complexity of the systems [13]. The security of the data is highly important because the leakage of medical data to any malicious user could manipulate the data and this manipulation severely affects the treatment of the individual, sometimes this disrupted treatment results in the death of the individual [24]. In smart cities, the medical data vulnerability is preserved by the encryption of data, which denies access to the third party and securely transmits the data with high authentication [2]. The cloud-based system could be initiated, processed, and terminated based on the application due to the characteristics of flexibility [28]. The confidentiality of the data is enhanced by the encryption technique that provides access to multiple individuals with the key attributes and on the receiver side, the data gets decrypted [14].

Due to the increased mortality rate, the researchers analyzed and executed multiple diagnosis methods for the diagnosis of heart diseases [21]. Deep learning algorithms, which are frequently employed for heart disease prediction and which are crucial for the exact and accurate diagnosis of coronary heart disease [26], collect prior knowledge and information from related pathological occurrences. The monitoring of heart disease patients should be based on all aspects of the body parameters that include blood pressure, gender, diabetes, and various other habitual actions. The quality of the service in the medical field is enhanced by these artificial intelligence techniques and the facilities are promoted even in the remote areas [12]. The cost of the system is also reduced [4]. The CNNs in deep learning techniques highly focus on the necessary attributes and eliminate redundant, irrelevant data and improve the quality of prediction [27]. The employed deep learning techniques are also useful in the processing of various other diseases, such as knee osteoarthritis, brain tumor classification, speech recognition, and various other applications [26].

A smart heart disease prediction model using cloud computing is introduced together with an effective smart security solution for medical data. A secure heart disease prediction model is performed using the ESVO-based Swish Bessel classifier and the contributional significance is the utility of the ESVO algorithm that helps in the tuning of the hyper parameters present in the classifier, and the utility of the Swish Bessel function allows better information flow for the classifier. This helps in attaining better prediction of heart disease and the data in the cloud server is also authenticated and authorized using specified techniques. Many existing models are developed for secure heart disease prediction but contain several security issues and take more time to predict. By utilizing our proposed model several benefits occur such as high security is provided in this model for the medical data, hence fraudsters can’t able to access the data, and takes less time to predict. In this research, the data collected are stored in the cloud and the security is ensured by proper authentication and authorization, hence is called a secure heart disease prediction model. The contribution of the study is as follows,

ESVO optimization: The ESVO optimization is developed by the hybridization of Emperor Penguin optimization (EPO) [3] and African Vulture Optimization Algorithm (AVOA) [1] which helps in the tuning of the weight and bias parameters in the classifier by improvising the speed and gaining deeper information.

ESVO-based Swish Bessel CNN classifier: The CNN classifier predicts heart disease effectively without human intervention improves the prediction rate and minimizes the error occurrence. The utility of the Swish Bessel function helps in the coherent information flow.

Authorization and authentication: The authorization is performed using the salt digital algorithm and the authentication is performed using the EAN-13 algorithm that effectively authorizes and authenticates the user and the owner while performing the transmission of data.

The research flow of the manuscript is as follows: Section 2 lists the existing methods utilized for the detection of heart disease along with the methods pros and cons. Section 3 presents the IoT network’s system concept, and Section 4 presents the secure heart disease prediction model’s full detailed methodology using the proposed ESVO-based Swish Bessel CNN classifier. Section 5 offers an interpretation of the findings produced by the proposed methodology, and Section 6 offers a conclusion to the study.

2. Motivation

Heart disease is an deadliest diseases and it must be identified as soon as possible to decrease the death count. Hence secure heart disease prediction is performed in this research which faces several security concerns in safeguarding the data and predicting the disease with high accuracy. Below are some state-of-the-art methods that can be used to get a better understanding of secure heart disease prediction.

2.1. Literature review

The literature works carried out by the previous researchers are enumerated as follows: Simanta Shekhar Sarmah [22] executed a patient monitoring scheme using a modified deep learning network that authenticated, encrypted, and classified the heart disease data. The authentication is performed using the SH-2512 and the disease is classified as normal and abnormal. The method utilized lesser time for the process of encryption but the accuracy of the system needs improvement. M. Vedaraj and P. Ezhumalai [26] introduced homomorphic encryption for ensuring privacy and is integrated with cryptography techniques that secured data privacy and predicted heart disease with high accuracy and less cost but the robustness of the system is not evaluated. Sundara Velrani Karuppiah and Geetha Gurunathan [13] executed a fish and whale algorithm for improving the security and the Support Vector Machine (SVM) is availed for the classification of the disease. This method provided security in both the personal and public domain but when the performance of the system is affected when there is a presence of noise. Haedar Al-Safi and Jorge Munilla [2] utilized two layers, one layer consists of blockchain for security purposes and the other layer consists of the records related to the medical data for the diagnosis process. The utility of the optimization algorithm for the selection of features helped to optimize the irrelevant data. The occurrence of error is reduced but the number of transactions is limited during the particular time. Chuan Zhang et al. [29] secured the patient’s historical data and stored in the cloud storage and the model is trained using a learning algorithm based on perceptron that preserved the privacy of data with less cost but the efficiency of preserving the data could be further improved. Ling Li et al. [14] achieved attribute revocation and improved the efficiency of the cipher text update. This method reduced the computational complexity and increased the memory capability by performing the process in the fog nodes. The method is secured under the Decisional Bilinear Diffie–Hellman (DBDH) assumption. The application of blockchain technology could furthermore enhance the revocation process. Yuanyuan Pan et al. [17] enhanced the deep learning availing the CNN and detected the heart disease with an improved processing time, which helped the doctors to predict the disease based on the information in the cloud at any place. The misclassification of the method is considerably low but the method should be improved considering the feature selection. Aqsa Rahim et al. [19] dwelled with the imbalanced data and missing values using SMOTE and performed the diagnosis of heart disease using a machine learning-based framework. This method reduced the problems arouses due to overfitting or underfitting but the algorithm used here struggles to maintain the threshold values that are predefined.

A secure heart disease prediction model that utilizes an ESVO-based Swish Bessel CNN classifier has been designed to address security concerns and improve the security of individuals. Authentication and authorization of data act as a challenge in most of the methods and here the process is performed in the cloud, where the data collected from the networks is stored there then access requests are sent to the cloud server and if the availability of data is needed, after the process of verification, the authenticated and genuine user receives the data that greatly reduced the chance of data breaches, which also decreased time delays.

2.2. Problem statement

Whenever there is a need for data, the cloud server receives an access request, and following the necessary verification, the verified and authenticated user is given access to the necessary data. However, safeguarding the privacy of each individual is a difficult challenge. Heart disease prediction with the highest accuracy is also a challenging problem in the research.

2.3. Challenges

The challenges to be performed in the research are as follows,

The blocks in the blockchain perform a limited amount of transactions at a particular time, along with that a delay is provided for maintaining the security of the data that slows down the process of transmission, and sometimes the blockchain network initiates the need for large storage space for the storing of data posses various challenges [2].

The disease diagnosis should be performed through the medical centers but the information about the patient should not get revealed due to privacy issues. Maintaining the privacy of the individual is also a challenging task [26].

Attaining effective interoperability is a strenuous task since the data are stored in the cloud in IoT devices and the problem is faced when the cloud system is integrated with the healthcare systems and the accessibility is made through public and local clouds [19].

The transmission of medical data is vulnerable to various attacks and maintaining security in a threatening environment is a challenging task [13].

Maintaining healthcare records and attaining the integrity of the data is a challenging task to be performed [22].

3. System model for IoT network

The network model of the IoT consists of doctors, patients, cloud servers, and IoT sensors. The IoT networks are highly suitable for large-scale systems since it consists of several cloud servers. The IoT system consists of several fields and the field comprised a reasonable number of patients. The distributed fields could be accessed through the cloud server. In the case of a real-time scenario, if the doctor needs to access the data from the patients through the IoT nodes implanted in the patient several authentication schemes should be followed between the doctor, patient, and the IoT nodes for security purposes. The authentication is carried out in three phases i) authentication between the doctor and the cloud server ii) authentication between the cloud server and the patient iii) authentication between the patient and the doctor. For providing secure communication between the doctor and the patient, a session key is generated for authentication purposes for futuristic purposes. The system administrator initiates the required credentials needed for the authentication purpose through the registration phase and stores the credentials in the smart card, then share the essential credentials with the patient as well as doctors. This could be useful in monitoring the patients even in quarantine conditions without physical contact. Let us consider m number of IoT nodes, which are equipped for patients in the distributed environment. The IoT nodes are denoted as, $\begin{matrix} (1) & I = {I_{1}, I_{2}, I_{3}, \dots I_{i} \dots I_{m}} \end{matrix}$

Each node $I_{i}$ engages in collecting heart-related data from the patients. Since the IoT nodes are battery-operated, the energy efficiency of the IoT nodes is assured to ensure prolonged data communication between the patient and the hospital portal. The collected data is stored in the hospital portal, from which the disease classification is done using the CNN classifier, which assures an efficient diagnosis for the patient. For storing the medical data of the patient in the portal, patient authentication is required for which a cipher text-based BAR code is used. The system model of the IoT network used in this research is shown in Fig. 1.

Fig. 1.

System model of the IoT network.

4. Secure heart disease prediction model availing ESVO-based Swish Bessel CNN classifier

The main aim of the research is to predict heart disease from the data collected using IoT networks. The medical data in the IoT networks are highly vulnerable to various threats hence the data present in the networks should be highly authenticated. The data collected from the networks are stored in the cloud and the authentication and authorization of data are performed in the cloud. Whenever there is a necessity for data the access request is provided to the cloud server and after the verification process the authenticated and genuine user receives the data, which greatly avoids data breaches. The authentication is performed using the EAN-13 by generating a cipher text-based barcode and the authorization is performed using the salt-based digital signature. The security of the data is also enhanced using the EDH-ECC algorithm. The data collected from the cloud is preprocessed using the ADASYN algorithm for reducing the data imbalance problems and the necessary features are selected and then the classification is performed using the Swish Bessel – Convolutional Neural Network classifier. The schematic representation is shown in Fig. 2.

Fig. 2.

Schematic representation of the heart disease prediction model.

4.1. Enhanced security using cloud server

Generally, the data needed for the heart disease prediction from the patient is stored in the cloud server due to its high flexibility and security and in this research, the patient identification, authentication, authorization, and security are enhanced using specified techniques that are interpreted in below sections.

4.1.1. Patient identification and authentication using EAN-13

The patient portal authentication is performed to ensure the appropriate identification of the patient, in which the patient’s health information is stored. In the cloud server, the data of the patient is stored and before storing the authentication is performed. The authentication is performed to verify the genuine users and here the process is carried out efficiently using the image processing algorithm EAN-13. EAN-13 stands for European Article Numbering which is widely used for patient identification. The constitution of the barcode consists of 13 numeric characters represented in digits. The initial two or three digits designate the identity code of the patient, and the remaining numbers represent the respective hospital and the code of the departments in the hospital. The final digit is used as the checksum that verifies the data for the occurrence of errors. The EAN-13 barcode shows a 13-digit code graphically as a line pattern. The EAN-13 barcode is made up of a pattern that alternates between different-sized black (1) and white (0) bars. One could imagine that such a line pattern is created using dark bars and light sections with varying thicknesses (width-1, 2, 3, or 4).

4.1.2. Two-level authentication

The System entities used for the authentication are the user $U_{r}$ , cloud server, and patient $I_{i}$ . Initially for accessing the data, the login request from the user is received and is denoted as, $\begin{matrix} (2) & L = {R_{r}, T_{1}, d_{1}, d_{2}, d_{3}} \end{matrix}$ here, $R_{r}$ designates the dynamic factor of the users, $T_{1}$ denotes the time stamp, $d_{1}$ , $d_{2}$ , $d_{3}$ and denotes the security factors. Now the cloud server validates the timestamp $T_{1}$ as, $\begin{matrix} (3) & | T_{1} - T_{1}^{l} | ⩽ Δ T \end{matrix}$

Here, the receiving time of the data attributes l is represented by $T_{1}^{l}$ and if the above condition is satisfied then the communication is blocked. Further, the cloud server validates the database for the valid dynamic factor ${FR}_{r}$ and temporal dynamic factor ${tR}_{r}$ to find the dynamic factor $R_{r}$ , when there is no match then the session ends up invalid $R_{r}$ . On the other hand, the cloud server receives the user information $U_{r}$ concerning $R_{r}$ from the database. When $R_{r} = {FR}_{r}$ then the login is not successful. Now the server evaluates the message as, $\begin{array}{c} (4) & λ_{1} = H [R_{r} ‖ U_{r} ‖ T_{1}] \\ (5) & R_{r} = d_{1} \oplus λ_{1} \end{array}$

Then the cloud server receives the information from $I_{i}^{th}$ node concerning $R_{r}^{i}$ from the database. Now, the validation between the user $U_{r}$ and $i^{th}$ patient is done based on the authorization parameter F. If the user is not authentic then the access is not authorized. Otherwise, the user $U_{r}$ is authenticated by the cloud server. Then the gateway or cloud server updates the security factors. Accordingly, a random time stamp $T_{2}$ is generated, where the security factors are evaluated as, $\begin{array}{c} (6) & d_{5} = H [{FR}_{r} ‖ T_{2} ‖ R_{r}] \oplus w_{3} \\ (7) & d_{6} = λ_{6} \oplus H [{FR}_{r} ‖ T_{2}] \\ (8) & d_{7} = H [{FR}_{r} ‖ R_{r} ‖ H [{FR}_{r} ‖ λ_{1}]] \end{array}$

Now, the digital signature is generated, $M_{2} = {d_{5}, d_{6}, d_{7}, T_{2}}$ is communicated to $I_{i}$ through a secure public communication channel. When $M_{2}$ is received at $I_{i}$ , the timestamp $T_{2}$ is verified. If $| T_{2} - T_{2}^{'} | ⩽ Δ T_{2}$ then the session terminates. The attribute $T_{2}$ represents the actual time stamp, $T_{2}^{'}$ designates the time stamp of the message $M_{2}$ , $Δ T_{2}$ designates the preset transmission delay.

4.1.3. Authorization

The authorization process is used to revoke access permission for the registered user. To secure sensitive information and system resources, a range of authorization approaches are employed as access control solutions to secure, that only authorized parties have access to specified resources. Using access policies and tokens, the system administrator defines the actions that a user is permitted to perform on a particular IoT device. Generally, authorization strategies that use policy-based or token-based architectures are used often and have both benefits and drawbacks. The policy-based architecture is frequently employed in centralized systems that require a single entity to serve as an evaluation server and carry out access control policy. Permission-based on policy may lead to undesirable delays and higher communication costs. While policy-based designs are better suited for centralized systems, token-based architectures hold improvement as a viable alternative. It doesn’t require any more communication and does not affect system resources. There is a natural security issue in present applications because only authorized individuals can access certain IoT sensors.

To update the access control of the user $U_{r}$ , the system administrator chooses the ID of $U_{r}$ and using the secret key the system administrator evaluates the digital signature as, $\begin{matrix} (9) & PI = H [{IU}_{r} ‖ S_{k}] \end{matrix}$ here, ${IU}_{r}$ is the ID of the user and $S_{k}$ denotes the secret key.

The system administrator updates the authorization parameter $F_{U_{r}}^{*}$ of user $U_{r}$ to enable the data access to the user $U_{r}$ . Now, the system administrator sends the authorization intimation to the corresponding server through a public communication channel. The authorization message contains ${{IU}_{r}, PI, F_{U_{r}}^{*}}$ . The authorization message is then encrypted using a cloud server’s secret key. This encrypted authorization message is sent to all the cloud servers for authorizing the user for granting data access. Once the message has been received at the cloud server, it is decrypted using the cloud server’s secret keys, where the gateway acquires the user’s confidential information, and the new authorization parameter is generated as $F_{U_{r}}^{new}$ for the user for enabling the future login within the server.

4.1.4. EDH-ECC for data security

Security of the data is enhanced by the encryption of data using the EDH-ECC encryption mechanism and is stored in the cloud for further proceeding. A shared secret can be established between two parties using the Diffie–Hellman (ECDH) key agreement protocol across an unsecured channel if each party has an elliptic-curve public-private key pair. This shared information can either be utilized directly as a key or create another key. Subsequent communications can subsequently be encrypted with a symmetric-key cipher utilizing the key, or the key that was obtained from it.

Let us assume $G_{p}$ a finite field of order p and k is an additive group at a point M on a nonsingular elliptic curve $ξ (G_{p})$ of order p. A hashing function is selected as the symmetric encryption and is described by $ξ (\cdot)$ and therefore the public key is generated as follows, $\begin{matrix} (10) & P = S_{k} * p \end{matrix}$ here, P denotes the public key, $S_{k}$ designates the secret key, and p denotes the order of the elliptic curve. The elliptic curve equation is represented as, $\begin{matrix} (11) & ξ (G_{p}) = Z^{3} + q x + r \end{matrix}$

Before beginning the encryption process, the input data and its entropy are created. The EDH-ECC algorithm is then used to carry out the encryption. The input and entropy of the input data are expressed mathematically as follows, $\begin{matrix} (12) & D_{i}^{e} = {a_{1}, a_{2}, \dots, a_{i}^{j}, \dots, a_{i}^{m}, e (a_{1}), e (a_{2}), \dots, e (a_{i}^{j}), \dots, e (a_{i}^{m})} \end{matrix}$

4.2. Read the input data

The input data is collected from the patient affected by the heart disease and the input data for the $i^{th}$ patient is given by, $\begin{array}{c} (13) & D = {D_{1}, D_{2}, D_{3} \dots D_{i}, \dots D_{m}} \\ (14) & D_{i} = {a_{1}, a_{2}, \dots a_{i}^{j}, \dots a_{i}^{n}} \end{array}$ here, n denotes the total attribute in the $i^{th}$ data.

4.3. Pre-processing of data

In the medical field, generally, the data will be high in noise, and redundant and there will be the presence of irrelevant features with imbalance problems. For the effective prediction of heart disease, there is a necessity for clean, precise, and balanced data, which is attained through the preprocessing steps. In this research, preprocessing is performed using the ADASYN sampling approach. The ADASYN sampling approach is utilized because the imbalance problems are greatly reduced using this ADASYN approach. The ADASYN approach generates the minority data by interpreting the distribution of the samples and these synthetic minority data are difficult to learn. The ADASYN approach effectively reduces the learning bias introduced by the original imbalanced dataset along with that adaptively alters the decision boundary to focus on the samples that are the most difficult to learn.

4.4. Heart disease prediction using ESVO-based Swish Bessel CNN classifier

The CNN classifier is commonly used for prediction problems due to its automatic identification of features and prediction with high accuracy. To learn the complex patterns of the classifier, the Swish Bessel activation function is used and the classifier is tuned using the ESVO algorithm. The architecture of the ESVO-based Swish Bessel CNN classifier is shown in Fig. 3. The parameters are detailed in Table 1.

Fig. 3.

Architecture of ESVO-Swish Bessel CNN.

Table 1

Experimental setup of parameters

Layer (type)	Output shape	Param#
conv2d	(None, 2824, 1, 264)	4488
leaky_re_lu (LeakyReLU)	(None, 2824, 1, 264)	0
Maxpooling2D	(None, 942, 1, 264)	0
Conv2D	(None, 942, 1, 128)	304256
leaky_re_lu_1 (LeakyReLU)	(None, 942, 1, 128)	0
MaxPooling2D	(None, 314, 1, 128)	0
Conv2D	(None, 314, 1, 64)	73792
leaky_re_lu_2 (LeakyReLU)	(None, 32)	0
(MaxPooling2D)	(None, 105, 1, 64)	0
flatten (Flatten)	(None, 6720)	0
dense (Dense)	(None, 32)	215072
leaky_re_lu_3 (LeakyReLU)	(None, 32)	0
Dropout	(None, 2)	99

4.4.1. Convolutional layer

The Convolutional layer’s essential function is the extraction of relevant features that are necessary to identify the heart disease from the input in the form of feature maps. Several filters in the Convolutional layer grasp the information from the parameters throughout the learning phase. The enhancement of the edges and the sharpening of the blur images are also performed in the convolutional layer that is mathematically expressed as, $\begin{matrix} (15) & {(W_{λ}^{g})}_{Z, W} = {(B_{λ}^{g} (bias))}_{Z, W} + \sum_{s_{1} = 1}^{u_{1}^{s - 1}} \sum_{s_{2} = - u_{1}^{f}}^{u_{2}^{f}} \sum_{s_{3} = - u_{2}^{f}}^{u_{2}^{f}} {(W_{λ, s}^{f})}_{Z, W} * {(W_{s}^{f - 1})}_{Z + α, W + β} \end{matrix}$ here, ${(W_{λ}^{g})}_{Z, W}$ denotes the feature map given by the $f^{th}$ Convolutional layer centered at Z, W. The weight and bias of the layer are given by $W_{λ, s}^{f}$ and $B_{λ}^{g} (bias)$ are tuned using the EVSO optimization. The parameters $s_{1}$ , $s_{2}$ , $s_{3}$ denote the individual output of the Convolutional layers.

Proposed Emperor Spheniscidae Vampire Optimization algorithm

The Emperor Spheniscidae Vampire optimization algorithm reduces the complexity of the classifier by improving the efficiency and robustness of the classifier and the detailed interpretation of the algorithm is described in the below sections.

Motivation: Emperor Spheniscidae (Emperor penguin) is the heaviest and tallest species of penguin. Male and female Emperor Spheniscidae are extremely similar in terms of size and plumage. Emperor Spheniscidae has a black dorsal side, a head that is white, pale yellow breasts, and bright yellow ear patches. Emperor Spheniscidae lives in open ice and lays eggs throughout winter. Emperor Spheniscidae gathers in enormous colonies that number in the hundreds of thousands during the breeding season. Females can travel 50 kilometers to hunt in the ocean after laying a single egg. Emperor Spheniscidae is a gregarious species that engage in cooperative foraging and hunting. Emperor Spheniscidae is capable of diving to a depth of 1,900 feet. Spend at least 25 minutes underwater. The largest penguin species is the emperor penguin. It lacks flight and has feathered wings that have been stiffened and flattened, like other penguin species. Emperor Spheniscidae is the only animal that gathers together to stay warm during the Antarctic winter. Vampire relies on dead animals for their food and here the African vulture is considered as the vampire and the vampire possesses the flight behavior in the form of rotational behavior.

Mathematical model for the Emperor Vampire Spheniscidae optimization: The Emperor Spheniscidae’s huddling behavior served as the basis for the mathematical modeling of the optimization. (Emperor Penguin) and this optimization focuses on determining the effective mover. The shape of the huddling behavior is given by the L-shaped polygon plane and as an initial step, the Emperor Spheniscidae set up the boundary for the huddling phase, and then the computation of the temperature profile is performed. After that, the distance between the Emperor Spheniscidae is evaluated, which helps in determining the exploration and exploiting phase, and finally, the best mover in the Emperor Spheniscidae is determined.

Initialization of huddle boundary: During the huddling behavior, the Emperor Spheniscidae assigns the position in the shape of a polygon shape, and the wind flows around the huddle is estimated to locate the polygon’s huddle border. Emperor Spheniscidae moves at a velocity slower than the wind flow, which is constant. The huddle boundary generated randomly by Emperor Spheniscidae is derived from the hypothesis of complex variables. The velocity of the wind is described by the attribute $V_{wind}$ and the gradient of the function is given by ϕ and the gradient is given by, $\begin{matrix} (16) & ϕ = \nabla V_{wind} \end{matrix}$

The complex potential is generated by the combination of the vector λ and is given by, $\begin{matrix} (17) & Q = ϕ + i λ \end{matrix}$ here, the imaginary constant is described by i, Q denotes the analytical function of the polygon plane.

Temperature profile around the huddle: Emperor Spheniscidae huddles to conserve energy and keep the huddle at a comfortable temperature. An assumption is made as when the radius of the polygon assigns the value lesser than 1, then the temperature tends to 0 and vice versa. The exploration and exploitation of the Emperor Spheniscidae solely relies on the temperature profile and is computed as follows, $\begin{array}{c} (18) & H^{'} = (H - \frac{t_{max}}{t - t_{max}}) \\ (19) & H = \{\begin{array}{cc} 0; & if rad > 1 \\ 1; & if rad < 1 \end{array}\} \end{array}$ here, $H^{'}$ designates the temperature profile and H denotes the temperature. The number of iterations is designated by t and the maximum number of iterations is given by $t_{max}$ .

Interspacing of Emperor Spheniscidae: Succeeding the generation of the huddle boundary, the distance between the best candidate solution and the Emperor Spheniscidae is evaluated. The fittest solution is determined by the fitness function that presents closer to the candidate solution. The other candidate solution of Emperor Spheniscidae updates their positions concerning the best candidate solution and is mathematically expressed as, $\begin{matrix} (20) & A_{ES} = diff (B (\vec{C}) \cdot \vec{E} (t) - \vec{J} \cdot \vec{E_{ES}} (t)) \end{matrix}$

The best candidate solution and the Emperor Spheniscidae are separated by a symbol $A_{ES}$ , diff denotes the absolute difference among the Emperor Spheniscidae and the best candidate solution. The finest and fittest candidate solution of the Emperor Spheniscidae during the current iteration is given by $\vec{E}$ , and the position of this best Emperor Spheniscidae is given by $\vec{E_{ES}}$ . The collision between the neighbors is avoided by the vectors $\vec{C}$ and $\vec{J}$ and is computed as follows, $\begin{array}{c} (21) & \vec{C} = (U \times (H^{'}) + E_{grid}^{acc} \times rand ()) - H^{'} \\ (22) & \vec{J} = rand () \end{array}$ here, U denotes the movement parameter that helps in the avoidance of collision by maintaining the gap between the Emperor Spheniscidae, and the value of this moving parameter is set to 2. The temperature profile under the huddle is designated by $H^{'}$ and the attribute $E_{grid}^{acc}$ designates the accuracy of the polygon grid and is obtained by evaluating the difference between the individual Emperor Spheniscidae. A random function in the range $[0, 2]$ is given by $rand ()$ .

The factor $B ()$ is a parameter that initiates the social force to attain the best candidate solution and is evaluated using the formula, $\begin{matrix} (23) & B (\vec{C}) = {(\sqrt{a \cdot e^{- t / b} - e^{- t}})}^{2} \end{matrix}$ where the attributes a and b are control parameters and are used for attaining better exploration and exploitation results. The value of a is in the range $[2, 3]$ and the value of b is in the range $[1.5, 2]$ and the results are best in the provided ranges.

Reposition of the moving Emperor Spheniscidae: Relying upon the best optimal solution the position of the Emperor Spheniscidae gets an update and the moving Emperor Spheniscidae is responsible for the varying positions of the Emperor Spheniscidae and the vacating of the current position. The next position of the Emperor Spheniscidae is given by, $\begin{matrix} (24) & \vec{E^{ES}} (t + 1) = \vec{E} (t) - \vec{C} \cdot \vec{P_{ES}} \end{matrix}$ here, $\vec{E^{ES}} (t + 1)$ designates the updated position of the Emperor Spheniscidae, and once the mover has been shifted, the Emperor Spheniscidae huddling behavior is recalculated during the iteration phase.

Flight behavior of Vampire: The vampire exhibits the characteristics of rotational flight and is expressed by the spiral motion model. All the vampires possess rotational flight behavior and the best two vampires are selected as the optimal solution and is given by, $\begin{array}{c} (25) & {Vam}_{1} = E_{t}^{best} * (\frac{rand * E_{t}^{vam}}{2 π}) * cos (E_{t}^{vam}) \\ (26) & {Vam}_{2} = E_{t}^{best} * (\frac{rand * E_{t}^{vam}}{2 π}) * sin (E_{t}^{vam}) \\ (27) & E^{vam} (t + 1) = E_{cur}^{best} (1 - (\frac{rand * E_{t}^{vam}}{2 π}) * cos (E_{t}^{vam}) - (\frac{rand * E_{t}^{vam}}{2 π}) * sin (E_{t}^{vam})) \end{array}$ here, the best two vampires are given by ${vam}_{1}$ and ${vam}_{2}$ during the next iteration $E_{t + 1}$ and $E_{t}^{vam}$ denotes the current iteration of the vampire. sin and cos denotes the sine and cosine functions and $rand$ is the random number in the range $[0, 1]$ .

Inventive phase: When the rotational flight behavior of the vampire is coagulated with the position of the Emperor Spheniscidae then the best and finest solution could be obtained since the flight behavior helps in preventing the attack from the predators and this behavior enhances the foraging of food and the spiral rotation helps in the coverage of the larger area and helps in gaining a deeper insight of the area, that helps in attaining more information. The coagulated behavior is mathematically expressed by the following equation described below, $\begin{array}{c} (28) & E^{best} = 0.5 (E^{ES} (t + 1) + E^{vam} (t + 1)) \\ (29) & E^{best} = [\vec{E} (t) - \vec{C} \cdot \vec{P_{ES}}] + E_{cur}^{best} (1 - (\frac{rand * E_{t}^{vam}}{2 π}) * cos (E_{t}^{vam}) - (\frac{rand * E_{t}^{vam}}{2 π}) * sin (E_{t}^{vam})) \end{array}$

The Pseudocode for the ESVO algorithm is shown in Table 2. The flow diagram of the ESVO algorithm is displayed in Fig. 4.

Table 2
Pseudo code for the ESVO

S. no Pseudo code for the ESVO

1 Input: $D_{i}$

2 Output: $E_{best}$

3 Initialize huddling boundary

4 Huddling behavior

5 Determine gradient function: $ϕ = \nabla V_{wind}$

6 Generate complex potential: $Q = ϕ + i λ$

7 Temperature profile

8 Determine the temperature profile: $H^{'}$

9 Evaluate distance: $A_{ES}$

10 Collision avoidance: $\vec{C}$ , $\vec{J}$

11 Attain best candidate solution: $B (\vec{C})$

12 Evaluate repositioning behavior: $\vec{E^{ES}} (t + 1)$

13 Flight Behavior of Vampire

14 $E^{vam} (t + 1)$

15 Inventive phase

16 $E^{best} = 0.5 (E^{ES} (t + 1) + E^{vam} (t + 1))$

S. no	Pseudo code for the ESVO
1	Input: $D_{i}$
2	Output: $E_{best}$
3	Initialize huddling boundary
4	Huddling behavior
5	Determine gradient function: $ϕ = \nabla V_{wind}$
6	Generate complex potential: $Q = ϕ + i λ$
7	Temperature profile
8	Determine the temperature profile: $H^{'}$
9	Evaluate distance: $A_{ES}$
10	Collision avoidance: $\vec{C}$ , $\vec{J}$
11	Attain best candidate solution: $B (\vec{C})$
12	Evaluate repositioning behavior: $\vec{E^{ES}} (t + 1)$
13	Flight Behavior of Vampire
14	$E^{vam} (t + 1)$
15	Inventive phase
16	$E^{best} = 0.5 (E^{ES} (t + 1) + E^{vam} (t + 1))$

Fig. 4.

Flow diagram of the proposed model.

4.4.2. Activation function

The activation function generally helps the classifier to learn complex patterns by evaluating the weighted total and adding the bias. The Bessel–Swish activation function highly improves the information propagation and is given by, $\begin{matrix} (30) & ψ_{swish} (h) = h \cdot sigmoid (h) \end{matrix}$

4.4.3. Max pooling layer

This layer reduces the complexity of the network by performing the down-sampling operation. The pooling layer reduces the feature map size and the resultant of the pooling layer gives pooled feature maps.

4.4.4. Fully connected layer

The weights are assigned to the input vectors and the classification is performed using this layer. The final output is generalized and obtained from this fully connected layer and is given by, $\begin{matrix} (31) & FCL = δ {(W_{λ}^{g})}_{Z, W} = {(B_{λ}^{g} (bias))}_{Z, W} + \sum_{s_{1} = 1}^{u_{1}^{s - 1}} \sum_{s_{2} = - u_{1}^{f}}^{u_{2}^{f}} \sum_{s_{3} = - u_{2}^{f}}^{u_{2}^{f}} {(W_{λ, s}^{f})}_{Z, W} * {(W_{s}^{f - 1})}_{Z + α, W + β} \end{matrix}$

5. Results and discussion

This section includes a description of classifier performance in disease classification along with other related methods. The important parameters are taken into consideration for analyzing the Performance of the ESVO-Swish Bessel CNN approach for identifying heart disease, hepatitis, lung cancer, and Parkinson’s disease.

5.1. Experimental setup

The ESVO-based Swish Bessel CNN classifier is implemented in the PYTHON tool in the Windows 10 OS with 8 GB RAM to reveal the classification performance of the proposed as well as the existing methods. The evaluation is performed using the accuracy of the metric, which proves the overall significance of the class, sensitivity that provided the number of positive instances, and the negative instances is measured using specificity.

5.2. Dataset description

Heart disease database [ 7 ]: All the experiments are published using only 14 subsets, although there are 76 attributes available in the database. The machine learning-related researchers most significantly preferred the Cleveland database. The database mainly refers to analyzing patients having heart disease. The integer value varies from 0 to 4, in which the presence of disease is described as 1,2,3,4, and the absence is denoted as 0.

Hepatitis [ 8 ]: In the hepatitis database, there 19 attributes are present for the classification of diseases with 155 instances. The characteristics of the attributes are categorical, integer, and real, in addition to the dataset characteristics as multivariate.

Lung cancer [ 16 ]:

Three types of pathological lung cancers are described in this data with 56 attributes and 32 instances for the classification task. The characteristics of the attributes and the dataset is integer and multivariate.

Parkinson’s disease [ 18 ]:

This dataset consists of 31 people’s biological voice measures, 23 of whom have Parkinson’s disease (PD). Each column in the table similar to a particular voice measure, and each row in the table represents one of the 195 voice recordings from these individuals. The major objective of the data is to identify between healthy people and people with PD depending on the “status” column, which is set to 1 for PD and 0 for healthy. In a CSV file if one recording is present in each row enery patient can hold six recordings and first column presents the patients name.

5.3. Comparative methods

ANN [10], SVM [5], Random Forest [9], KNN [20], Deep CNN [6], Bi-LSTM [15], AVOA-Swish Bessel CNN [1], EPO-based Swish Bessel CNN [3] are the various existing methods considered for comparing the performance with the ESVO-Swish Bessel CNN classifier in accurate disease classification.

5.3.1. Comparative analysis for heart disease database based on training percentage

The ESVO-based Swish Bessel CNN classifier performance is evaluated by considering the performance measures with the other existing comparative methods based on the TP, which is revealed in Fig. 5.

Figure 5a) reveals the accuracy of the ESVO-Swish Bessel CNN classifier in classifying heart disease. The accuracy of the ESVO-Swish Bessel CNN classifier is 94.877 % for the TP 90 with a performance improvement of 1.69 % than the existing EPO-Swish Bessel CNN classifier.

Figure 5b) depicts the sensitivity of the ESVO-Swish Bessel CNN classifier in classifying heart disease. The sensitivity of the ESVO-Swish Bessel CNN classifier is 95.464 % for the TP 90 with a performance improvement of 7.72 % than the existing EPO-Swish Bessel CNN classifier.

Figure 5c) shows the specificity of the ESVO-Swish Bessel CNN classifier in classifying heart disease. The specificity of the ESVO-Swish Bessel CNN classifier is 93.293 % for the TP 90 with a performance improvement of 2.72 % than the existing EPO-Swish Bessel CNN classifier.

Figure 5d) shows the F1 measure of the ESVO-Swish Bessel CNN classifier in classifying heart disease. The F1 measure of the ESVO-Swish Bessel CNN classifier is 95.140 % for the TP 90 with a performance improvement of 1.49 % than the existing EPO-Swish Bessel CNN classifier.

Thus, the performance of the ESVO-Swish Bessel CNN classifier is better than the existing methods for the heart disease classification by well-tuning the hyperparameters of the classifier by the ESCO optimization techniques.

Fig. 5.

Comparative analysis for heart disease classification a) accuracy b) sensitivity c) specificity d) F1 measure.

Fig. 5.

(Continued.)

5.3.2. Comparative analysis for hepatitis database based on training percentage

Figure 6a) depicts the accuracy of the ESVO-Swish Bessel CNN classifier in classifying hepatitis disease. The accuracy of the ESVO-Swish Bessel CNN classifier is 94.178 % for the TP 90 with a performance improvement of 4.29 % than the existing EPO-Swish Bessel CNN classifier.

Figure 6b) shows the sensitivity of the ESVO-Swish Bessel CNN classifier in classifying hepatitis disease. The sensitivity of the ESVO-Swish Bessel CNN classifier is 94.079 % for the TP 90 with a performance improvement of 3.29 % than the existing EPO-Swish Bessel CNN classifier.

Fig. 6.

Comparative analysis for hepatitis disease classification a) accuracy b) sensitivity c) specificity d) F1 measure.

Fig. 6.

(Continued.)

Figure 6c) depicts the specificity of the ESVO-Swish Bessel CNN classifier in classifying hepatitis disease. The specificity of the ESVO-Swish Bessel CNN classifier is 93.004 % for the TP 90 with a performance improvement of 1.90 % than the existing EPO-Swish Bessel CNN classifier.

Figure 6d) shows the F1 measure of the ESVO-Swish Bessel CNN classifier in classifying hepatitis disease. The F1 measure of the ESVO-Swish Bessel CNN classifier is 94.642 % for the TP 90 with a performance improvement of 1.87 % than the existing EPO-Swish Bessel CNN classifier.

Thus, the performance of the ESVO-based Swish Bessel CNN classifier is better than the existing methods for the hepatitis disease classification by well-tuning the hyperparameters of the classifier by the ESVO-Swish Bessel optimization techniques.

5.3.3. Comparative analysis for lung cancer database on TP

Figure 7a) shows the accuracy of the ESVO-Swish Bessel CNN classifier in classifying lung cancer disease. The accuracy of the ESVO-Swish Bessel CNN classifier is 95.584 % for the TP 90 with a performance improvement of 0.47 % than the existing EPO-Swish Bessel CNN classifier.

Figure 7b) depicts the sensitivity of the ESVO-Swish Bessel CNN classifier in classifying lung cancer disease. The sensitivity of the ESVO-Swish Bessel CNN classifier is 96.279 % for the TP 90 with a performance improvement of 3.47 % than the existing EPO-Swish Bessel CNN classifier.

Fig. 7.

Comparative analysis for lung cancer disease classification a) accuracy b) sensitivity c) specificity d) F1 measure.

Fig. 7.

(Continued.)

Figure 7c) shows the specificity of the ESVO-Swish Bessel CNN classifier in classifying lung cancer disease. The specificity of the ESVO-Swish Bessel CNN classifier is 94.249 % for the TP 90 with a performance improvement of 0.47 % than the existing EPO-Swish Bessel CNN classifier.

Figure 7d) depicts the F1 measure of the ESVO-Swish Bessel CNN classifier in classifying lung cancer disease. The F1 measure of the ESVO-Swish Bessel CNN classifier is 95.584 % for the TP 90 with a performance improvement of 0.47 % than the existing EPO-Swish Bessel CNN classifier.

Thus, the performance of the ESVO-Swish Bessel CNN classifier is better than the existing methods for lung cancer disease classification by well-tuning the hyperparameters of the classifier by the ESVO-Swish Bessel optimization techniques.

5.3.4. Comparative analysis for Parkinson’s disease database on TP

Figure 8a) shows the accuracy of the ESVO-Swish Bessel CNN classifier in classifying Parkinson’s disease. The accuracy of the ESVO-Swish Bessel CNN classifier is 95.279 % for the TP 90 with a performance improvement of 1.13 % than the existing EPO-Swish Bessel CNN classifier.

Fig. 8.

Comparative analysis for Parkinson’s disease classification a) accuracy b) sensitivity c) specificity d) F1 measure.

Fig. 8.

(Continued.)

Figure 8b) depicts the sensitivity of the ESVO-Swish Bessel CNN classifier in classifying Parkinson’s disease. The sensitivity of the ESVO-Swish Bessel CNN classifier is 95.279 % for the TP 90 with a performance improvement of 0.44 % than the existing EPO-Swish Bessel CNN.

Figure 8c) shows the specificity of the ESVO-Swish Bessel CNN classifier in classifying Parkinson’s disease. The specificity of the ESVO-based Swish Bessel CNN classifier is 95.279 % for the TP 90 with a performance improvement of 0.88 % than the existing EPO-based Swish Bessel CNN classifier.

Figure 8d) depicts the F1 measure of the ESVO-Swish Bessel CNN classifier in classifying Parkinson’s disease. The F1 measure of the ESVO-Swish Bessel CNN classifier is 95.279 % for the TP 90 with a performance improvement of 0.89 % than the existing EPO-based Swish Bessel CNN classifier.

Thus, the performance of the ESVO-based Swish Bessel CNN is better than the existing methods for Parkinson’s disease classification by well-tuning the hyperparameters by the optimization techniques.

Table 3

Comparative table for the proposed and existing methods in disease classification

Database	Measures (%)	Training percentage 90

		Methods

		ANN	SVM	Random forest	KNN	Deep CNN	BiLSTM	AVOA-based Swish Bessel CNN	EPO-based Swish Bessel CNN	ESVO-based Swish Bessel CNN
Heart disease database	Accuracy	73.588	83.786	87.896	86.622	92.946	92.382	93.777	93.273	94.877
	Sensitivity	80.521	80.573	84.93	82.275	85.783	85.174	91.279	88.089	95.464
	Specificity	82.174	85.673	89.257	87.346	90.602	89.939	91.604	90.758	93.293
	F1 measure	77.608	77.901	86.837	84.343	89.256	88.188	94.697	93.724	95.14
Hepatitis database	Accuracy	73.267	85.187	88.233	89.153	89.488	89.602	92.624	90.138	94.178
	Sensitivity	81.776	87.481	88.167	89.154	90.773	90.811	92.122	90.983	94.079
	Specificity	76.405	82.907	88.09	88.512	88.561	88.934	91.89	91.241	93.004
	F1 measure	84.057	87.777	88.232	88.934	89.163	89.359	93.13	92.875	94.642
Lung cancer database	Accuracy	77.036	81.109	78.885	81.333	83.04	82.163	85.96	95.136	95.584
	Sensitivity	78.834	81.056	79.559	81.28	82.675	82.11	92.454	92.937	96.279
	Specificity	78.97	82.306	81.195	83.34	86.047	84.649	90.621	93.807	94.249
	F1 measure	76.412	81.934	78.605	82.562	85.111	84.853	86.944	95.136	95.584
Parkinson’s disease database	Accuracy	84.962	90.462	90.428	91.252	92.752	91.665	93.625	94.206	95.279
	Sensitivity	84.904	91.486	91.296	92.129	92.752	92.545	93.378	94.858	95.279
	Specificity	85.152	87.579	87.103	88.378	92.068	88.777	92.752	94.437	95.279
	F1 measure	87.251	91.692	91.092	91.923	92.752	92.338	93.904	94.434	95.279

5.4. Comparative discussion

The classification accuracy of the ESVO-based Swish Bessel CNN classifier for the heart, hepatitis, lung cancer, and Parkinson’s disease database is revealed in Table 3. The accuracy, sensitivity, specificity, and F1 measure of the ESVO-based Swish Bessel CNN classifier in heart disease classification is 94.877 %, 95.464 %, 93.293 %, and 95.14 % the ESVO-based Swish Bessel CNN classifier in hepatitis disease classification is 94.178 %, 94.079 %, 93.004 %, and 94.642 % the ESVO-based Swish Bessel CNN classifier in lung cancer disease classification is 95.584 %, 96.279 %, 94.249 %, and 95.584 % and the ESVO-based Swish Bessel CNN classifier in Parkinson’s disease classification is 95.279 %, 95.279 %, 95.279 %, and 95.279 %. The main challenge of the research is to optimize the classifier so that the time consumption, the convergence could be boosted. The optimization is performed using the ESVO algorithm, where the training time is reduced by the optimal tuning of the classifier. The computational complexity of thr proposed method along with the existing methods are analyzed and tabulated in Table 4. Further the parameters utilized in the proposed research are depicted in Table 5.

Table 4
Analysis of computational complexity

Methods Time consumption (ms)

ANN 81

SVM 78

Random forest 72

KNN 69

Deep CNN 64

Bi-LSTM 61

AVOA-based Swish Bessel CNN 58

EPO-based Swish Bessel CNN 55

Proposed ESVO-based Swish Bessel CNN 50

Methods	Time consumption (ms)
ANN	81
SVM	78
Random forest	72
KNN	69
Deep CNN	64
Bi-LSTM	61
AVOA-based Swish Bessel CNN	58
EPO-based Swish Bessel CNN	55
Proposed ESVO-based Swish Bessel CNN	50

Table 5

Parameters description

Variables	Description	Variables	Description
$R_{r}$	Dynamic factor of the users	$T_{1}^{l}$	Receiving time of the data attributes
$T_{1}$	Time stamp	l	Data attributes
$d_{1}$ , $d_{2}$ , $d_{3}$	Security factors	${FR}_{r}$	Valid dynamic factor
${tR}_{r}$	Temporal dynamic factor	$U_{r}$	User
$I_{i}$	Patient	F	Authorization parameter
$T_{2}$	Actual time stamp	$T_{2}^{'}$	Time stamp of the message
$Δ T_{2}$	Preset transmission delay.	$S_{k}$	Secret key
${IU}_{r}$	ID of the user	$F_{U_{r}}^{*}$	Authorization parameter
$F_{U_{r}}^{new}$	New authorization parameter	$G_{p}$	Finite field of order p
k	Additive group at a point M	$ξ (G_{p})$	Nonsingular elliptic curve
$ξ (\cdot)$	Symmetric encryption	P	Public key
$S_{k}$	Secret key	p	Order of the elliptic curve
n	Total attribute in the $i^{th}$ data	${(W_{λ}^{g})}_{Z, W}$	Feature map given by the $f^{th}$ Convolutional layer cantered at Z, W.
$W_{λ, s}^{f}$	Weight	$s_{1}$ , $s_{2}$ , $s_{3}$	Individual output of the Convolutional layers
$V_{wind}$	Velocity of the wind	ϕ	Gradient of the function
i	Imaginary constant	Q	Analytical function of the polygon plane.
$H^{'}$	Temperature profile	H	Temperature
t	Number of iterations	$t_{max}$	Maximum number of iterations
$A_{ES}$	Distance between the Emperor Spheniscidae and the best candidate solution	diff	Absolute difference between the Emperor Spheniscidae and the best candidate solution
$\vec{E}$	Finest and fittest candidate solution of the Emperor Spheniscidae during the current iteration	$\vec{E_{ES}}$	Position of this best Emperor Spheniscidae
U	Movement parameter	$H^{'}$	Temperature profile under the huddle
$E_{grid}^{acc}$	Accuracy of the polygon grid	$B ()$	Parameter
a, b	Control parameters	$\vec{E^{ES}} (t + 1)$	Updated position of the Emperor Spheniscidae
${vam}_{1}$ & ${vam}_{2}$	Best two vampires	$E_{t + 1}$	Next iteration
$E_{t}^{vam}$	Current iteration of the vampire	sin, cos	Sine and cosine functions
$rand$	Denotes the random number in the range $[0, 1]$ .	$B_{λ}^{g} (bias)$	Bias

6. Conclusion

A secure heart disease prediction system is developed by using an ESVO-based Swish Bessel CNN classifier, with high efficiency and security. Due to the sensitivity of the medical data, the authentication and authorization of data are carried out in the cloud. The usage of the cloud provides high data security, assists collaboration, enhances storage capacity, high mobility, and low maintenance cost. The early prediction of heart disease greatly helps in preventing life-threatening individuals by taking preventive measures at an earlier time. The usage of the ESVO-based Swish Bessel CNN classifier plays a significant role in heart disease prediction because of the incorporation of the ESVO algorithm that effectively tuned the parameters. The information propagation of the classifier is also boosted using the Swish Bessel activation function which also helped in improving the accuracy of the prediction. The applications of such analysis include Healthcare information systems, personal health records, E-health and telemedicine, clinical decision system, cloud-based digital libraries, and Drug discovery. The efficiency of the research is proved by measuring the accuracy, sensitivity, specificity, and F1-measure, which obtained the values of 94.877 %, 95.464 %, 93.293 %, and 95.14 % which are more efficient than the previous state of art methods. A model’s performance is monitored and evaluated using metrics throughout training and testing. However, the performance metric can also be utilized as a loss function if it is differentiable for specific tasks.

References

Abdollahzadeh,

F.S.

Gharehchopogh and

Mirjalili, African vultures optimization algorithm: A new nature-inspired metaheuristic algorithm for global optimization problems, Computers & Industrial Engineering 158 (2021), 107408. doi:10.1016/j.cie.2021.107408.

Al-Safi,

Munilla and

Rahebi, Patient privacy in smart cities by blockchain technology and feature selection with Harris Hawks Optimization (HHO) algorithm and machine learning, Multimedia Tools and Applications 81(6) (2022), 8719–8743. doi:10.1007/s11042-022-12164-z.

S.K.

Baliarsingh,

Ding,

Vipsita and

Bakshi, A memetic algorithm using emperor penguin and social engineering optimization for medical data classification, Applied Soft Computing 85 (2019), 105773. doi:10.1016/j.asoc.2019.105773.

Bhambere,

Abhishek and

Sumit, Rapid digitization of healthcare – a review of COVID-19 impact on our health systems, Int. J. All Res. Educ. Sci. Methods 9 (2021), 1457–1459.

Bhatia,

Prakash and

G.N.

Pillai, SVM based decision support system for heart disease classification with integer-coded genetic algorithm to select critical features, in: Proceedings of the World Congress on Engineering and Computer Science, 2008, pp. 34–38.

N.I.

Hasan and

Bhattacharjee, Deep learning approach to cardiovascular disease classification employing modified ECG signal from empirical mode decomposition, Biomedical signal processing and control 52 (2019), 128–140. doi:10.1016/j.bspc.2019.04.005.

Heart disease database, https://archive.ics.uci.edu/ml/datasets/heart+disease.

Hepatitis database, https://archive.ics.uci.edu/ml/datasets/hepatitis.

Javeed,

Zhou,

Yongjian,

Qasim,

Noor and

Nour, An intelligent learning system based on random search algorithm and optimized random forest model for improved heart disease detection, IEEE Access 7 (2019), 180235–180243. doi:10.1109/ACCESS.2019.2952107.

10.

Jeyaranjani,

T.D.

Rajkumar and

T.A.

Kumar, Coronary heart disease diagnosis using the efficient ANN model, Materials Today: Proceedings (2021).

11.

Jiwani,

Gupta and

Afreen, A convolutional neural network approach for diabetic retinopathy classification, in: 2022 IEEE 11th International Conference on Communication Systems and Network Technologies (CSNT), IEEE, 2022, pp. 357–361. doi:10.1109/CSNT54456.2022.9787577.

12.

Jiwani,

Gupta and

Whig, Novel HealthCare framework for cardiac arrest with the application of AI using ANN, in: 2021 5th International Conference on Information Systems and Computer Networks (ISCON), IEEE, 2021, pp. 1–5.

13.

S.V.

Karuppiah and

Gurunathan, Secured storage and disease prediction of E-health data in cloud, Journal of Ambient Intelligence and Humanized Computing 12(6) (2021), 6295–6306. doi:10.1007/s12652-020-02205-6.

14.

Li,

Wang and

Li, Efficient attribute-based encryption outsourcing scheme with user and attribute revocation for fog-enabled IoT, IEEE Access 8 (2020), 176738–176749. doi:10.1109/ACCESS.2020.3025140.

15.

Li,

Zhang,

Dai,

Zhou and

Wang, Interpretability analysis of heartbeat classification based on heartbeat activity’s global sequence features and BiLSTM-attention neural network, IEEE Access 7 (2019), 109870–109883. doi:10.1109/ACCESS.2019.2933473.

16.

Lung cancer database, https://archive.ics.uci.edu/ml/datasets/lung+cancer.

17.

Pan,

Fu,

Cheng,

Tao and

Guo, Enhanced deep learning assisted convolutional neural network for heart disease prediction on the Internet of medical things platform, IEEE Access 8 (2020), 189503–189512. doi:10.1109/ACCESS.2020.3026214.

18.

Parkinson’s database, https://archive.ics.uci.edu/ml/datasets/parkinsons.

19.

Rahim,

Rasheed,

Azam,

M.W.

Anwar,

M.A.

Rahim and

A.W.

Muzaffar, An integrated machine learning framework for effective prediction of cardiovascular diseases, IEEE Access 9 (2021), 106575–106588. doi:10.1109/ACCESS.2021.3098688.

20.

Rajathi and

Radhamani, Prediction and analysis of rheumatic heart disease using kNN classification with ACO, in: 2016 International Conference on Data Mining and Advanced Computing (SAPIENCE), IEEE, 2016, pp. 68–73.

21.

Sahija, Critical review of mixed reality integration with medical devices for patientcare, 2022, p. 10.

22.

S.S.

Sarmah, An efficient IoT-based patient monitoring and heart disease prediction system using deep learning modified neural network, IEEE Access 8 (2020), 135784–135797. doi:10.1109/ACCESS.2020.3007561.

23.

S.N.

Tadiboina, The use of AI in advanced medical imaging, Journal of Positive School Psychology 6(11) (2022), 1939–1946.

24.

S.N.

Tadiboina and

Kumar, Treatment policies for chronic illnesses and the potential to transform health care with artificial intelligence, Ilkogretim Online 18(1) (2019), 610–619.

25.

S.N.

Tadiboina and

Liu, Artificial intelligence (AI) and deep learning (DL) in medical diagnosis process such as SPECT and PET, Journal of Positive School Psychology 6(8) (2022), 10665–10673.

26.

Vedaraj and

Ezhumalai, HERDE-MSNB: A predictive security architecture for IoT health cloud system, Journal of Ambient Intelligence and Humanized Computing 12(7) (2021), 7333–7342. doi:10.1007/s12652-020-02408-x.

27.

Whig,

Gupta and

Jiwani, Real-time detection of cardiac arrest using deep learning, in: AI-Enabled Multiple-Criteria Decision-Making Approaches for Healthcare Management, IGI Global, 2022, pp. 1–25.

28.

Whig,

Gupta,

Jiwani,

Kouser and

Anand, Adaptive clinical treatments and reinforcement learning for automatic disease diagnosis, in: AI-Enabled Multiple-Criteria Decision-Making Approaches for Healthcare Management, IGI Global, 2022, pp. 204–221. doi:10.4018/978-1-6684-4405-4.ch011.

29.

Zhang,

Zhu,

Xu and

Lu, PPDP: An efficient and privacy-preserving disease prediction scheme in cloud-based e-healthcare system, in: Future Generation Computer Systems, Vol. 79, 2018, pp. 16–25.

Secure heart disease prediction model using ESVO-based Swish Bessel CNN classifier

Abstract

Keywords

1. Introduction

2. Motivation

2.1. Literature review

2.2. Problem statement

2.3. Challenges

3. System model for IoT network

4.1.1. Patient identification and authentication using EAN-13

4.1.2. Two-level authentication

4.1.3. Authorization

4.1.4. EDH-ECC for data security

4.2. Read the input data

4.3. Pre-processing of data

4.4. Heart disease prediction using ESVO-based Swish Bessel CNN classifier

4.4.3. Max pooling layer

4.4.4. Fully connected layer

5. Results and discussion

5.1. Experimental setup

5.2. Dataset description

5.3. Comparative methods

5.3.1. Comparative analysis for heart disease database based on training percentage

Table 4 Analysis of computational complexity Methods Time consumption (ms) ANN 81 SVM 78 Random forest 72 KNN 69 Deep CNN 64 Bi-LSTM 61 AVOA-based Swish Bessel CNN 58 EPO-based Swish Bessel CNN 55 Proposed ESVO-based Swish Bessel CNN 50

References

Table 4
Analysis of computational complexity

Methods Time consumption (ms)

ANN 81

SVM 78

Random forest 72

KNN 69

Deep CNN 64

Bi-LSTM 61

AVOA-based Swish Bessel CNN 58

EPO-based Swish Bessel CNN 55

Proposed ESVO-based Swish Bessel CNN 50