1 Introduction

Data perturbation is commonly used and approved by the Privacy Data Mining technique, which implicitly requires data miners to have single-level trust [1, 2]. In addition, before the data sharing process, the individual authenticated procedure has been performed [3, 4]. In a multilayer trust environment, data miners with a wide trust level can access fewer perturbed copies [5, 6]. Moreover, at the lower trust phase, these less disrupted copies are inaccessible to data miners [7]. Also, in some cases, data miners with a wide trust range might have access to alter copies at multiple trust levels [8, 9]. Additionally, data miners with varying levels of trust may collaborate to exchange perturbed documents [10, 11]. With the growing demand for privacy protection problems, the Machine Learning (ML) based privacy-preserving model has garnered considerable interest from academics and industry [12, 13]. But, the majority of existing approaches have practical limitations [14]. The fundamentals of securing medical cloud data are described in Fig. 1.

Fig. 1
figure 1

Medical cloud fundamentals for preserving the privacy

Full size image

In addition, most cryptographic techniques are provably secure, which imposes significant computational and transmission overhead [15, 16]. The trustworthiness of the relatively effective privacy-preserving approaches is being contested due to their inability to be proven secure [17]. In recent years, neural processes have been widely applied in various real-world contexts [18, 19]. The Healthcare system is the foundation for health-associated decision-making and comprises four critical functions [20]: data compilation, creation, synthesis, and analysis and usage [21, 22]. The Healthcare monitoring sector has collected information from the clouds sensed by IoT gadgets [23, 24]. Moreover, securing the cloud data is more important to protect the data against malicious events [25]. In addition, the privacy score of the designed model has been analyzed by launching malicious activities in the developed, designed model [26, 27]. While launching the attack, if the confidential range of the channel has remained in a constant wide range, then it’s sufficient for protecting the cloud data [28, 29]. Besides, if the privacy range has been decreased when the attack is in the transmit layer, then there is possible to get an attack [30, 31]. A decent ML scheme frequently requires data from numerous sources [32]. However, data resources often reticently distribute their data because it contains users' personal information. As a result, determining the training model for private info is significantly difficult. Several works have been implemented, such as fuzzy neural systems [33], Apache systems [34], etc., to end these issues. But, the appropriate results are not found. So, an optimized ML system has been planned to maintain cloud data privacy.

The Honey Pot-based Modular Neural System is a proposed solution to preserve privacy in medical data. The system is designed to protect sensitive medical data from unauthorized access or disclosure while allowing for the effective use of the data for research or other purposes. The system is based on a modular neural network architecture that uses honeypots to detect and deter unauthorized access attempts. The basic idea behind the system is to create a virtual environment that mimics the real medical data system but with fake data designed to attract hackers or other unauthorized users. These fake data sets are referred to as "honey pots." When an unauthorized user attempts to access the honey pot, the modular neural network is triggered, and a series of actions are initiated to detect and deter the intrusion attempt. The modular neural network consists of multiple modules, each performing a specific function. These modules include data acquisition, feature extraction, preprocessing, and classification. The data acquisition module collects medical data from various sources, such as electronic medical records or wearable devices. The feature extraction module identifies relevant features in the data that can be used to identify patterns or anomalies. The data preprocessing module cleans and normalizes the data to prepare it for analysis. The data classification module uses machine learning algorithms to classify the data and identify potential intrusions. In addition to the modular neural network, the system includes several other security features, such as access controls, encryption, and firewalls. These features protect real medical data from unauthorized access or disclosure. Overall, the Honey Pot-based Modular Neural System represents a promising approach to preserving privacy in medical data. The system can effectively detect and deter unauthorised access attempts by creating a virtual environment that mimics the real data system but with fake data designed to attract intruders. The modular neural network architecture and other security features provide additional layers of protection to ensure that sensitive medical data remains private and secure.

2 Related Works

A few recent associated works are described as follows:

Securing medical data in digital records is crucial for smart applications. So, Aqsa Mohiyuddin et al. [33] have designed a fuzzy neural system for the IoT sensed data to separate the normal and disease data from the sensed information. The separated data is stored separately in the cloud environment. The authentication conditions must be satisfied if anybody needs to retrieve the data. However, it has required more resources to execute.

The data stored in the cloud is unstructured, so maintaining privacy is complicated. Hence, the Apache system is utilized by Rashid et al. [34] for the real-time application to secure the stored data. Moreover, it has afforded the finest confidential range but needs more resources to run the process. In addition, the ML approaches are incorporated to identify the disease types and store the data based on disease types.

Gadekallu et al. [35] have designed a malicious forecasting system based on the block chain strategy for medical cloud data. The main purpose of this model is to predict the tampering activities in the stored data. Hence, the designed Blockchain model is tested with e-health applications and has recorded the best outcome. However, if the data size has been increased, it needs more time to execute, and the confidentiality score has been minimized.

Offering a better communication range towards the model is a big challenge. So, Babar et al. [36] have introduced the topology system incorporating communication parameters. Also, securing the cloud data demand-side approach has been employed and tested in the grid environment. Moreover, this model includes an agency system, which predicts the authentication process. But designing this system is very complex.

Cloud data is vulnerable to get an attack, so Golec et al. [37] have introduced a lightweight-based privacy design for the cloud environment to hide the stored data from third parties. In addition, based on the dataset, some modifications were required in the lightweight protocol. This model has provided high privacy for cloud data. But designing this approach has required more resources.

Masud et al. [38] introduced a low-weighted and physically secure mutual validation and protection key organization protocol that utilizes physical unclonable functions (PUFs) to ensure the system validates the user and sensor link before validating a session key. The proposed protocol provides all the required protection properties. And also utilizes low sources to manage and is secured from physical attacks. Hence it is more suitable for IoT-based medical system applications. Yet, it is dependent on the internet.

Nowadays, computer-based models like natural language and texture processing are applied to investigate sentimental assessment to detect and extract emotion. So, Alowibdi et al. [39] conducted a sentimental evaluation using a high-dimensional Twitter database of English tweets based on ten emotional themes. Experimental results show that COVID-19 has spread happiness and hope, fear, gratitude and mixed emotions among human beings for various reasons. However, the dataset consists of biased feelings.

In order to detect and recognize the COVID-19 case, Rahman et al. [40] developed a deep neural model. This research also studies federated learning in COVID-19 cases where edge learning can compute private datasets. Point-of-care ways can be combined with deep neural model-based RT-PCR lab examination outcomes to compare the multi-modality of COVID-19 identification and give more dimensions and diagnosis. This method provides good results. Yet, it needs continuous updating to increase accuracy.

Depending on deep learning, Sedik et al. [41] proposed a convolutional neural network (CNN) and convolutional long and short-term memory (ConLSTM). Two different types of databases are considered: CTges and X-ray images. This dataset consists of normal and COVID-19. Also, pneumonia and COVID-19 photos are categorized to validate the process. This method attained maximum accuracy and f-measure values for some cases. But, it is only accurate for some cases.

The world has continuously changed for the past two decades. The rapid evolution of technology is getting more issues. In order to tackle these problems, the united nation declared sustainable development goals (SDGs). It intended to stop hunger and poverty, AIDS and gender discrimination. So, SDGs should be investigated. Hence, Chopra et al. [42] studied about 17 SDGs in India’s 29 states. This research predicts a few countries like China, Germany, Japan, India, etc. Yet, many more countries are not considered.

Pashchenko, [43] investigates the transformation of 26 project teams from leading IT corporations to high-tech companies with strong internal practices like amazon, exness, VTB, BSC group, etc. Researchers estimate the outcomes of rapid adaption to transformations, work progress and predict for 2021. The main motive of this research work is to understand the new process in software development, hiring experts and organizing teams related to high-tech companies' refusal. Yet there is, technical issues occur.

Zhou et al. [44] proposed an effective technique called the filtering method for the COVID-19 scenario and identified the DDoS attack. This research proposed a statistical method, such as packet value and entropy changes for detection of DDoS attack traffic which is implemented in Omnet +  + and validated in various cases. And this proposed method effectively detects the DDoS attack traffic accurately from the flash crowd. However, the cost is high, and the design is complex.

In recent days, IoT has played a major role in every activity. It can be utilized in households to wireless networks. Anyhow, there are some limitations like security and shortcoming issues. So, Tewari et al. [45] discussed the problems in the ultra-low weight method because of its sensitivity to passive secret disclosure attacks and proposed a novel technique to tackle the shortcomings issues. The proposed model utilizes timestamps with the bitwise operation to protect from de-synchronisation and disclosure. The performance analysis is carried out to validate the proposed model. But, there is limited processing power.

Bulla et al. [46] proposed multiple agent-dependent datasets gathered, and aggregated method is presented to monitor fog computing architecture. The dataset collection method considers an integrated push–pull algorithm which enhances the data while changes in the new dataset are compared with the old dataset. In order to decrease communication overhead (CH) among cloud and fog nodes, a tree-dependent data aggregation method is designed. The developed model effectively improves the data coherency with lower CH than others. Yet, there is a high resource required.

With the rapid advancement of technology, IoT has become a significant part of people. There are so many security issues with novel techniques that make the backbone of the IoTs. A few effective security methods have been designed. The challenges are increasing, and the solutions have been improving. Adat et al. [47] investigated the history, background, IoT statistics, and IoT framework safety-based assessment. This paper discusses various research papers' techniques, advantages, drawbacks, and future scope.

Zhang et al. [48] proposed a secured de-centralized spatial crowdsourcing method for the 6G-based network in the box called DSC-NIB. The base station (BS) and sensor nodes can collect and transfer data on the blockchain by NIB without based upon the other party using the proposed DSC-NIB. The BS transmits encrypted location method metrics set for negotiating session keys and group keys with sensor nodes whose location satisfies the location method when validating the safety of the sensor node’s location. The novel technique enhances the performance by about 30 to 50% compared to others. But, there is complexity in designing.

The key contributions of the work are described as follows:

  • The IoT-based medical dataset (Disease symptom data) is initially gathered and imported into the python system.

  • Consequently, a novel HbMNS has been designed with suitable parameters to secure the stored data.

  • Initially, the preprocessing functions were activated to filter the noise from the trained database

  • Moreover, the authenticated module has been created to verify the original user.

  • The system's robustness is analyzed by launching the attack in the data transit layer.

  • Subsequently, the parameters are validated regarding the accuracy, Sensitivity, precision score, F-measure, Error, and run time.

3 System Model and Problem Statement

Cloud data privacy management is the hottest topic in all fields. Moreover, securing the medical data is crucial to preventing misunderstanding of the diagnosis report. So, the framework has been introduced for the cloud application to maintain the privacy of IoT-sensed data. But in many cases, securing the data processor has failed because of the attack vulnerabilities and harmfulness. These issues have motivated this research to implement a medical cloud data privacy model.

The system model and the problem statement are illustrated in Fig. 2. In the present model, the prediction and classification of diseases are incorrect due to attacks. Also, no continuous monitoring or verification process is available, making the system inaccurate. Hence, a novel HbMNS model is developed by integrating two effective model principles to secure the medical data and the verification process. The IoT-based medical database is taken from the standard website to validate the proposed model and is fed into the developed model's input phase. The dataset is preprocessed in the hidden layer in the proposed model to neglect the unwanted data. Then, the meaningful features are extracted and classified as a disease and no disease in the feature extraction module. And to check the security of the data, the attack is launched in the designed model. The proposed model efficiently detects the malicious and neglects it from the data. Thus, the presented model effectively provides disease classification and data privacy.

Fig. 2
figure 2

System model with the problem statement

Full size image

4 Proposed HbMNS for Securing Medical Data

The present research article has plans to implement a novel Honeypot-based Modular Neural System (HbMNS) introduced to secure the medical data stored in the cloud environment. Initially, the medical data was taken and trained to the system then a novel HbMNS was designed to secure the trained data. Moreover, the robustness of the developed model is analyzed by launching the attack in the cloud data transaction layer. Finally, the key metrics are calculated and compared with other models. The proposed architecture is described in Fig. 3.

Fig. 3
figure 3

HbMNS Framework

Full size image

The presented HbMNS model is a hybrid technique that merges two methods: the Honeypot Optimization algorithm [49] and the modular neural network. Moreover, the perturbation function is incorporated into the designed model to provide continuous monitoring. Initially, the collected dataset is trained and preprocessed to remove the noise features. The meaningful features are extracted by feature extraction, and diseases are classified in the classification layer. Then, a DoS attack is launched in the system to validate the system outcomes. At last, the medical data are secured and monitored continuously with a perturbation function. Thus, the system provides privacy to medical data and protects the data from attacks.

4.1 Layers of Designed HbMNS

In the designed HbMNS model, there are five different layers. The 1st layer is the input surface, where the collected dataset gets imported and initialized in the system. The secondary layer is hidden. Here the dataset is preprocessed to remove the noise features from the dataset.

The ternary layer is the classification layer; here, the present disease features in the dataset get extracted, and the diseases are classified according to symptoms. The fourth layer is the optimization layer, and here the perturbation function is incorporated to provide data privacy and continuous monitoring through the verification process. The final frame is the output layer. Here the outcomes of the designed model are estimated.

4.1.1 Preprocessing Phase

To design the proposed HbMNS, an input dataset containing parameters like diseases, symptoms, and severity is imported into the system. To initiate the process, the imported dataset must be initialized. The Initialization of the imported dataset is done based on a modular neural network (MNN) [50]. The dataset is initialized as in Eq. (1).

$$f(D_{\alpha ,\beta } ) = \zeta \left( {z_{1} ,z_{2} ,...z_{n} } \right)$$
(1)

where, \(D\) is the input medical dataset, \(\alpha\) and \(\beta\) are the dataset parameters, \(\zeta\) is the features present in the dataset and \(z_{n}\) is all data in the dataset. After initialization, the next step is preprocessing. Preprocessing removes the noise features from the input dataset. The preprocessing equation of the dataset is expressed in Eq. (2).

$$\rho (D) = \sum\limits_{i = 1}^{{z_{n} }} {D_{\alpha ,\beta } (} \zeta - \eta \nu )$$
(2)

where, \(\eta \nu\) indicates the noise features in the dataset and \(\rho\) denotes preprocessing variable.

4.1.2 Feature Extraction

Feature extraction is done to extract meaningful and meaningless features separately from the preprocessed dataset. Initially, the present features are tracked. The present features contain both meaningful and meaningless features. Here, the feature extraction is done based on Honeypot optimization. The feature tracking equation is expressed in Eq. (3).

$$F^{\prime}_{R} = \lambda \left( {\psi_{{\delta ,\,\,\,\delta^{*} }} } \right)$$
(3)

where, \(\psi\) is the present features containing both meaningful data \(\delta\) and meaningless features \(\delta^{*}\), \(\lambda\) is the feature tracking variable and \(F^{\prime}_{R}\) is the feature tracking function. The meaningless features are neglected to enhance the classification accuracy of the system.

$$\sigma_{g} = \lambda .\left( {2 \times \psi - \delta^{*} } \right)$$
(4)

where, \(\sigma_{g}\) indicates the extracted meaningful feature from the dataset. The removal of meaningless features is expressed in Eq. (4). The disease classification becomes more accurate by neglecting the meaningless features.

4.1.3 Disease Classification

After extracting the meaningful features, it is trained in such a way as to classify the diseases based on symptoms. The classification equation is expressed in Eq. (5).

$$G^{\prime}_{D} = \left\{ \begin{gathered} \,if(S_{m} = d_{Sm} )\,;\,\,d \hfill \\ else\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,;\,\overline{d}\, \hfill \\ \end{gathered} \right.$$
(5)

where, \(G^{\prime}\) is the disease classification variable, \(S_{m}\) refers to the symptom, \(d_{Sm}\) indicates the disease symptoms, \(d\) refers to disease and \(\overline{d}\) indicates no disease.

4.1.4 Perturbation Function

The perturbation function provides data privacy by hiding the dataset features with external data. It alters the original data in a particular format to protect it from attacks. Also, a verification module is added in Perturbation to provide continuous monitoring [51]. Here, the random noise addition perturbation approach provides privacy in medical data. The general equation for Perturbation is represented in Eq. (6).

$$\left[ {\hat{Z}} \right] = \,\left[ Z \right]\, + \,\left[ \varepsilon \right]$$
(6)

where Z-matrix is the original data, \(\varepsilon\)—matrix is the external data added to the original data for hiding, \(\hat{Z}\)—matrix is the new form of original data. The perturbation function is expressed in Eq. (7).

$$\left[ \begin{gathered} \hat{z}_{1,1} \,\,\,\,\hat{z}_{1,2} \,\,\,\,\hat{z}_{1,n} \,\,\,\, \hfill \\ \hat{z}_{2,1\,\,\,\,\,\,\,} \hat{z}_{2,2} \,\,\,\hat{z}_{2,n} \hfill \\ \hat{z}_{n,1} \,\,\,\,\,\hat{z}_{n,2} \,\,\,\hat{z}_{n,n} \hfill \\ \end{gathered} \right]\,\,\, = \,\,\,\left[ \begin{gathered} z_{1,1} \,\,z_{1,2} \,\,z_{1,n} \hfill \\ z_{2,1} \,\,z_{2,2} \,\,z_{2,n} \hfill \\ z_{n,1} \,\,z_{n,2} \,\,\,z_{n,n} \hfill \\ \end{gathered} \right]\,\, + \,\,\left[ \begin{gathered} \varepsilon_{1,1} \,\,\,\,\varepsilon_{1,2} \,\,\,\varepsilon_{1,n} \hfill \\ \varepsilon_{2,1} \,\,\,\,\varepsilon_{2,2} \,\,\,\varepsilon_{2,n} \hfill \\ \varepsilon_{n,1} \,\,\,\,\varepsilon_{n,2} \,\,\,\varepsilon_{n,n} \hfill \\ \end{gathered} \right]$$
(7)

Moreover, the system outcomes are validated by launching attacks. After Perturbation, the classification is accurate even after launching attacks, as the attackers cannot access the dataset. Moreover, data privacy and security are enhanced by incorporating a verification module. Two important constraints are required to access the dataset file in the verification process: filename and password. The user can access the file if the filename and password are valid Fig. 4.

Fig. 4
figure 4

Layers of HbMNS

Full size image
figure d

The functioning of the system is illustrated in pseudocode format in algorithm 1. The workflow of the designed HbMNS model is shown in Fig. 5. In the presented model, data privacy is verified by launching attacks before disease classification. If disease classification is accurate, then the system provides better outcomes.

Fig. 5
figure 5

Work flow of HbMNS

Full size image

5 Result and Discussion

The main objective of the HbMNS model is to protect medical data from attacks and to provide privacy to medical data. Here, the perturbation function is incorporated into the designed HbMNS model to give privacy to medical data. Also, the system's performance is validated by launching a DoS attack. Moreover, monitoring of the system is provided by the verification module. The implementation parameters description is listed in Table 1.

Table 1 Specification of implementation parameters
Full size table

Moreover, the working of the presented model is validated with a case study. In addition, the outcomes of the executed model are estimated before and after fixing the perturbation function. At last, the improvement score of the designed model is calculated with comparative analysis.

5.1 Case Study

The case study explains the working of the presented model. Here, the proposed method is validated with case (1) Disease classification before launching an attack, case (2) Disease classification after launching an attack, and case (3) Disease classification after Perturbation Table (2).

Table 2 Disease and symptom features
Full size table

Case 1: Disease classification before an attack.

The medical dataset containing the disease and symptoms features is initially imported into the system. Then, the imported dataset gets initialized like neurons in a modular neural network. After initialization, the input dataset is preprocessed to remove the error/ noise features. In the preprocessing phase, the noise features are removed; thus, the disease prediction can be accurate. After preprocessing, the next phase is feature extraction.

The present attributes in the preprocessed dataset are tracked and extracted in feature extraction. Then, the meaningless characteristics are neglected to enhance system performance. Therefore, the extracted dataset contains only the meaningful disease and symptoms characteristics. After feature extraction, the dataset is trained in such a way as to classify the disease under the symptoms. Before launching the attack, the disease classification is accurate. The system's performance before launching the attack is shown in Fig. 6.

Fig. 6
figure 6

Performance of the system before an attack

Full size image

Case 2: Disease classification after an attack.

After training the dataset, a DoS attack is launched to substantiate the system's performance. The system provides inaccurate classification after launching a DoS attack. For example, the system must classify the disease as a fungal infection according to symptoms such as itching, skin rash, and dyschromia patches. But the system classifies the disease as an allergy (inaccurate classification).

The system cannot provide accurate classification because the attacker changes the dataset attributes. The system's performance after launching an attack is shown in Fig. 7. After instigating the attack, data privacy and performance are minimized. Hence, the system's performance after launching an attack is lower than before.

Fig. 7
figure 7

Performance of the system after an attack

Full size image

Case 3: Disease classification after Perturbation.

In this case, the disease classification is estimated after incorporating the perturbation function. In the perturbation function, the filename and password are changed to the third form by multiplying them with other attributes without altering the original ones. The perturbation function protects the dataset and provides monitoring with a verification block. Thus, the dataset cannot be accessed unless the filename and password are known.

The results of the designed model after the Perturbation are shown in Fig. 8. Perturbation prevents attackers in the system and enhances system security by providing privacy to medical data. Thus, the system's performance after fixing the Perturbation is high. Also, it gives accurate disease classification.

Fig. 8
figure 8

Performance of the HbMNS model after Perturbation

Full size image

The overall performance of the designed model is displayed in Fig. 9. The outcomes are enhanced after incorporating the perturbation function into the proposed design. After Perturbation, the users cannot access the file without a filename and password. Thus the privacy of data is improved in the designed model.

Fig. 9
figure 9

Overall performance Analysis of the HbMNS model

Full size image

5.2 Comparative Analysis

A comparative assessment was made to determine the improvement score and manifest the better results of the designed model. The proposed HbMNS was validated with launching the attacks. As we expect, the proposed model effectively detects and removes the malicious. Here, the performance of the developed HbMNS model in prediction and neglecting the attack is compared with different models like Support Vector Machine (SVM) [52], K-star classification method (K-star CM) [53], Spatial Clustering of Applications Based on Density along with Noise (SCABDN) [34] and Multiple Channel Spatial–temporal Convolutional Neural Network (MCSCNN) [54] regarding the precision, f-measure, recall and accuracy.

F-measure: F-score is determined for estimating the data privacy provided by the system. Generally, the f-measure is calculated from the recall score and precision value. The equation for f-score estimation is expressed in Eq. (8).

$$F - measure\,\,\,\,\, = 2 \times \left[ {\frac{{R_{S} \times P_{V} }}{{R_{S} + P_{V} }}} \right]$$
(8)

where, \(R_{S}\) indicates the recall of the system and \(P_{V}\) refers to the precision value.

Recall- Score: Recall-score is estimated to measure the system performance by counting the true and false negatives of all positives attained. The recall-score equation is expressed in Eq. (9).

$${\text{Recall}}\,\,\,{\text{Score}}\,\,\,\, = \frac{{X_{p} }}{{X_{p} + Y_{n} }}$$
(9)

Here, \(X_{p}\) and \(Y_{n}\) refers to true and false negatives, respectively.

The comparison of the f-score and recall metric of the designed model with various models is shown in Fig. 10. Normally, the f-measure and recall of the system are calculated to estimate the system performance. The proposed model attained a high f-score and recall of 99.17% and 99.11%, respectively. At the same time, the f-score obtained by exiting models like SVM, SCABDN, and K-star CM is 98.4%, 95.78% and 93.99%, respectively. Similarly, the recall-score achieved by different models are 97.8%, 85.97%, and 95%, which is low compared to the proposed method.

Fig. 10
figure 10

Comparison of F-score and recall metric

Full size image

Precision: Precision is estimated to measure the system's accuracy by classifying the true and false positives out of all positive values attained. The precision equation is expressed in Eq. (10).

$${\text{Precision}}\,\,\,\% \,\,\,\, = \left[ {\frac{{X_{p} }}{{X_{p} + Y_{p} }}} \right] \times 100\,\,\,$$
(10)

where, \(X_{p}\) and \(Y_{p}\) refers to true and false positives, respectively.

The precision of the designed data privacy model with the existing model is shown in Fig. 11. The precision percentage achieved by the proposed model is 99.29%. But the precision obtained by the existing techniques like SCABDN, K-star CM, and MCSCNN is 96.63%, 94.5%, and 89.29%, respectively. The executed model attained high precision percentage than the existing designs.

Fig. 11
figure 11

Comparison of precision of HbMNS with existing models

Full size image

Accuracy: Accuracy is determined to determine the system's performance. It determines how efficiently the proposed model provides privacy to medical data. Accuracy is calculated in terms of true and false positives and negatives. The accuracy equation is expressed in Eq. (11).

$${\text{Accuracy}}\,\,\,\,\,\, = \frac{{X_{p} + X_{n} }}{{X_{p} + Y_{p} + X_{n} + Y_{n} }}$$
(11)

Here, \(X_{n}\) and \(Y_{n}\) defines true and false negatives, respectively.

The accuracy of the designed model is 99.18%, which is high compared to other designs. The accuracy achieved by other models like MCSCNN, K-star CN, SCABDN, and SVM is 95%, 95%, 96.81%, and 97%, respectively. The comparison of the accuracy of execution with other methods is shown in Fig. 12. The above analysis verified that the accuracy is enhanced in the presented model.

Fig. 12
figure 12

Comparison of Accuracy of different models

Full size image

The comparative evaluation of the designed HbMNS model with various techniques is tabulated in Table 3. From the comparative assessment, the executed model provided better accuracy, precision, recall, and f-measure than the other models.

Table 3 Comparative Analysis
Full size table

6 Discussion

A novel HbMNS model was proposed to protect the medical dataset from attacks and to provide privacy by hiding the data with a perturbation function. The Perturbation is incorporated into the executed model to give privacy to data with the verification module. The functioning of the system is explained in the case study. In addition, the system's outcomes are determined before and after incorporating perturbation fitness. In the end, the proposed model is validated with comparative analysis.

The evaluation of outcomes of the presented model is tabulated in Table 4. The outcomes of the executed model are evaluated through f-score, precision, recall, error rate, and accuracy. From comparative and performance evaluation, the results obtained by the presented model are high; that is, it achieved a high f-score of 99.17%, better accuracy of 99.18%, greater recall score of 99.11%, less error percentage of 0.82%, and high precision of 99.29%. Thus, the presented model provides better outcomes than other designs.

Table 4 Performance analysis of the HbMNS model
Full size table

7 Conclusion

To protect the medical database from attackers, a hybrid HbMNS design was developed. The executed model is validated with disease classification. Incorporating the perturbation function in the presented model provides better outcomes and classification accuracy. Also, the working of the designed system is explained with a case study in three different cases. The results of the system are evaluated before and after Perturbation. Moreover, the system's accuracy is validated by launching attacks on the system. In addition, the comparative valuation verifies that the presented model achieves greater accuracy of 99.18%. That is, the accuracy of the system is improved by 2.18%. Moreover, the precision and recall score of the system is enhanced by 2.59% and 1.31%, respectively. Also, the error rate is minimized in the executed model by 2.13%. Hence, the presented model protects the data from attacks and provides data privacy by hiding the data.