Users can access the system, which is intended for any vital information. The blockchain administrator collects and merges all dispersed keys, which the blockchain administrator only creates a key for safe key transfer33. If the user’s information matches, the corresponding user is added in the cloud storage. Otherwise, the process goes back to registration stage.
In this section, the Matrix-based Rivest-Shamir-Adleman (RSA) encryption method handles secure transmission of data and authorization, as depicted in the next sub-section. Various technologies like economic models, consensus algorithms, cryptography, and mathematics are present in the blockchain. Each record in the block model of a transactional database is saved using a secure distributed ledger. Both peer-to-peer networks and consensus mechanisms offer distributed data synchronization. The block is mostly made up of the header and data. As a result, the header contains the nonce, Merkle root, timestamp, current block, and preceding block information. Based on the Merkle tree, the main component of blockchain technology, the data content with its security authentication is accelerating. Pairing two transactions together and hashing them produces the hash. Each IoT node is validated in the IoT network using the Merkle tree without downloading and verifying the complete block. The Merkle root is stored in the block header, allowing each network node to confirm the transaction.
We used the Matrix-based RSA encryption method to handle secure transmission of data and authorization. While the idea of storing verified IoT data on the blockchain is appealing, its centralized nature often leaves it susceptible to external threats. As a result, the encryption approach is employed. This project handles storage using a decentralized interplanetary file system (IPFS). To construct the Merkle tree, data is encrypted and then partitioned into several blocks before being saved in the IPFS file system. The core idea behind this proposed approach to enhance IoT privacy is integrity verification, which makes use of matrix-based RSA encryption technology. This technique has the potential to detect tampering with the Internet of Things data stored in the blockchain database by either blockchain members or cloud servers. Each transaction reflecting data produced by IoT-based health monitoring equipment is cryptographically linked to previous entries and stored immutably using a decentralized ledger structure. This prevents unauthorized tampering or manipulation of the data. Access control restrictions may be enforced using smart contracts, ensuring that only authorized healthcare practitioners or stakeholders can access or make changes to sensitive medical records. In addition, data is protected before it reaches the blockchain via Matrix-based RSA encryption, which provides an extra strong level of anonymity throughout transmission and storage.
where in which x and y create two large prime numbers.From select invertible matrix I. Take and (K, A, I) as the private and public key from . For key generation, randomly selects the encryption matrix. The matrix A determines the number of data blocks. Even without the knowledge of determine the factor. For solving the smaller integer values are selected. Identify matrix in which s-fold application of exponentiation encrypts IoT data and is decrypted using
To prevent the disclosure of sensitive information, verify each value of . It is the responsibility of the encryptor to ascertain if the randomly chosen matrix is invertible.
The invertible matrix I encrypted with (A-1) different blocks (X1, X2,…., XA-1) for each new digital data block (Ym), which is given as follows:
Combine data from previous cycles with past encryptions. Encrypt and confound the encrypted data.
From the sender, receive n encrypted data blocks r every set. The matrix F calculates the decryption process.
Integrating the key generation center, cloud server, data owner, and third-party auditor (TPA) into blockchain-based data storage allows the Internet of Things (IoT) node to preserve its data. The IoT user sends their data to a blockchain-enabled cloud server. Following data storage, the user is required to sign a verification contract with the TPA. When it comes to data, the TPA is the one to trust. An affiliation between the cloud server and the cloud service provider (CSP) allows the CSP to offer storage services. It provides server power for processing and data storage. Giving back to the user is the main focus of the TPA, and it accomplishes this by communicating verification findings to the user and the server in the cloud. There is an instantaneous detection of data corruption. All interactions between the TPA and any entity are verified. An incomplete private key is generated by the Key Generation Center (KGC) under the jurisdiction of the authority. This center uses the user’s identification. The blockchain utilizes all four matrix-based RSA encryption parameters, which are listed below:
The dangers arise in the adversarial paradigm, which comprises dishonest auditors, malicious users, and partially trustworthy servers. Data theft can be concealed by a semi-trusted organization, such as a cloud server, by providing false evidence information to fool the TPA. However, while the cloud server may be able to modify the public key of an Internet of Things device, it will not have access to the master key of the KGC. Despite being able to replace its master key in some situations, the KGC is unable to access its public key. On rare occasions, the TPA might act inappropriately by delaying the completion of the verification. The compromised TPA could potentially crash into the cloud server, resulting in undetected data corruption. In certain cases, hostile Internet of Things (IoT) nodes can upload encrypted health data to a server in the cloud, which puts patients at risk. Figure 2 expresses the flowchart description of secure data communication using matrix-based encryption. A structural improvement over classic RSA, matrix-based RSA uses matrix operations to encode data blocks, allowing for increased encryption speed via parallelization of calculations. This technology enhances scalability in IoT applications without substantially increasing computing complexity or key length by enabling concurrently processing multi-dimensional data, such as time series, from wearable sensors. When hardware acceleration is available, such as GPU or SIMD support, matrix-based RSA may go faster than classical RSA since matrix operations are optimized for these architectures. Conversely, matrix-based RSA usually has greater computing overhead for key generation and encryption at the same security level as more modern algorithms like Elliptic Curve Cryptography (ECC) or lattice-based schemes, especially for devices with limited power at the edge. For instance, ECC is more efficient for large-scale IoT installations because it provides equal security with lower key sizes and quicker execution on restricted devices.
The data is safely retrieved and consolidated on the public cloud server. The health cloud’s structural design includes many patients’ health data. The detected values of the various patients were then downloaded to the Hospital’s specialized cloud server. As a result, the downloaded information is decrypted. The decryption technique makes use of the distributed key. The decryption process is the inverse of encryption, and the decrypted data is expressed as follows,
The m-number of decrypted values is signified as and the decrypted dataset is .
The proposed DCNN-based AO method to predict heart disease is discussed in this section. The section below describes the heart disease prediction model.
The several layers that make up a convolutional neural network (CNN) include the following: input, softmax, batch normalization, convolution, class output, and completely connected. The neural network’s first and input layer is responsible for receiving the raw data. Consistent with the input layer size, the data is input. The feature diversity is supplied by a variety of convolution kernels of slides, filters, or convolutional layer kernels, and different local information is recorded in this way. The values are chosen by the learners during the training process. It is the feature map that determines the padding, stride, filter size, and filter count.
A total of 100 neurons in present in the dense layer at the end of the architecture. The heart disease prediction using the DCNN-based AO algorithm is illustrated in Fig. 3. In many practical situations, the distribution of data in a dataset is not uniform. Most cases of cardiac disease detection have imbalanced data. The objective is to encourage more representation of underrepresented groups in designations.
The efficacy of accuracy (and error rate) in assessing a classifier’s performance is an evident difficulty that stems from the class imbalance problem. This is because most conventional classifiers focus on accuracy optimization and produce models similar to the naive model mentioned before. Although accurate, such a classifier is worthless in most real applications since the minority class is frequently the class of interest (otherwise, a classifier would not be required, as the class of interest occurs practically always). As a result, several strategies have been created to address the class imbalance issue. These techniques may be divided into two broad categories: sampling and skew-insensitive classifiers. The current machine learning approach becomes increasingly skewed toward the majority classes as the number of classes increases. The misclassification of minority classes occurs frequently. The DCNN-based AO model is utilized to solve this problem, and it is not biased toward the majority classes as the number of samples grows. The DCNN’s accuracy improves as the number of cases rises, not declines. Finally, the DCNN-based AO algorithm distinguishes between groups with typical and atypical heart disease. Based on this defined result, the hospital’s administration determines whether or not there is heart disease. If the data analyzed contains significant abnormalities, the HM sends an alarm message to the patient’s mobile phone.
Following pre-processing, feed the pre-processed data into the DCNN-AO algorithm for heart disease prediction. DCNN is an excellent classification approach that enhances accuracy. However, the increased number of layers, neurons, dropout rate, and other hyperparameters hampered the CNN’s performance. For DCNN hyperparameter optimization, we used the Archimedes Optimization (AO) algorithm in this study. The AO algorithm outperforms others in terms of convergence speed, searchability, the ability to exploit and explore, efficiency, and computational time, among other factors. The AO algorithm effectively optimizes the CNN hyperparameters in this study, resulting in improved classification results. Maximum accuracy is considered the fitness value. We looked at challenges like CNN hyperparameter tuning (HT) with improved prediction performance (CP), both multi-objective optimization problems.
Inside the parameter space, the design variable vector has as its upper boundary and as its lower boundary. Where U represents the fitness of the parameters, the vector representation of the objective space and the target function must be maximized. The hyperparameters are vital in improving the CNN’s performance and mainly depend on parameters such as regularization coefficient, number of epochs, momentum, etc. Table 2 displays the parameters that were optimized by the AO algorithm. An illustration of the AO algorithm’s flowchart for DCNN architecture optimization can be shown in Fig. 4. Neural networks use a regularization technique called dropout to prevent overfitting. A random subset of neurons is “dropped out” during training by having their outputs set to 0 with a certain frequency. Because it can no longer rely on certain neurons, the procedure encourages the network to acquire more reliable and independent features. Dropout enhances generalization while lowering the likelihood of overfitting. The Dropout in the model is given a value of 0.2 to avoid overfitting. The DCNN architecture is good at automatically finding important features from raw sensor inputs, such as ECG signals or vital signs. This allows for capturing spatial and temporal relationships crucial for accurate heart disease prediction. Hyperparameter optimization, which includes learning rates, convolutional filter sizes, and dropout ratios, is crucial to DCNN success. To solve this issue, we used the Archimedes Optimization method, which works like how things float in water to effectively search through hyperparameter options, helping us avoid getting stuck in less optimal solutions and speeding up the process.
