Computing, Communication and Learning

The expansion of remote-based digital healthcare-based IoT systems has accelerated the transfer of medical data through the IoT platform. This study proposes a unique model based on intelligent encryption algorithms in light of lightweight block ciphers, which will secure healthcare data transmitted via IoT devices to cloud systems; overall system is integrated with fog and edge computing to capture and process the data close to fog and edge devices. Compared to conventional encryption techniques, the suggested approach requires the least amount of time to generate cipher text information. The time complexity is decreased to 4.2 ms, and the power consumption is reduced to 7.97 mW, which also improves performance.

The rising usage of the cloud nowadays and its usage in various domains have made it more essential for all, which has led to an expansion in the size of data kept in the cloud. Data is the gold of our era; thus, it is important to protect it against any attacks. The intrusion detection system IDS is considered among of the most important solutions that address security issues and threats in the different models of cloud service delivery. IDS-based on machine learning (ML) has been developed to monitor and analyse data packets to detect abnormal behaviours and new attacks. The datasets utilized for these objectives are generally vast and include a lot of features, making computing very time-consuming. It is crucial to pick relevant features to include in the model, which produce better results and require less computation time than using all of the features. In this paper, we developed a system that combines filter-based feature selection with the support-vector-machine (SVM) model as a classifier. The NSL-KDD, Kyoto, and the CSE-CIC-IDS-2018 datasets are used to validate our system. We have compared with many existing methodologies and found that our proposed system outperformed the others in terms of accuracy, recall, precision, F-measure, and false-alarm rate.

The conceptual underpinnings of machine learning and artificial intelligence have been significantly impacted by deep learning, which has experienced substantial practical success. A deep learning model’s (DL) performance is influenced by a number of hyperparameters during model formation. The learning rate is one of them (LR). The examination of the widely used Adam optimizer’s LR impact on performance metrics and the choice of the best LR with no ambiguity for future research are the main objectives of this work. The DDoS SDN dataset and Long Short Term Memory (LSTM) DL model are used in the research to support the work for attack detection in fog-based IoT systems for security purposes. It was discovered that Adam Optimizers’ default LR 0.001 is the best; however, when huge batch sizes are taken into account, LR 0.01 proved to be better in the numbers of performance metrics but the noise is too high. As a result, it is decided that the LR 0.001 is the most appropriate value while building a DL model using the Adam optimizer.

Serverless computing has been widely used in several applications, where tasks are represented through an independent stateless functions. These stateless functions can be orchestrated in the form of workflow and can be deployed and executed in serverless frameworks. A number of serverless frameworks have emerged recently to provide orchestration services for serverless workflow. It is necessary to explore the potential of serverless frameworks which can be helpful for developers in taking business decisions. The performance of a serverless framework depends on several complexities associated with the serverless workflow. In this paper, we have focused on the effect of activity complexity associated with the workflow. The activity complexity of a workflow refers to the number of functions in the serverless application. A comparative analysis of various serverless workflow services based on change in performance parameters based on activity complexity is presented. The performance parameters such as overall function time, overhead time, and total response time are considered as the evaluation parameters. The extensive comparison has been done by considering both sequential and parallel workflow, which have been generated by various workflow services. Some of the findings are also presented for comparative analysis. This could be a potential resource for the application developers who have focused on serverless computing.

The blockchain is an emerging technology and is used in various applications for data security and trustworthiness. One fundamental technique of blockchain is that any changes incorporated by the authorized user in the system have been inserted into the transaction and would be traceable. The smart contract access control mechanisms have gained considerable attention since its applications. However, there are no systematic efforts to analyze existing empirical evidence. To this end, we aim to synthesize literature to understand the state-of-the-art of smart contract-driven access control mechanisms concerning underlying platforms, utilized blockchain properties, nature of the models, and associated test beds and tools. The attribute-based access control access rights are granted to users by evaluating suitable attributes. An essential aspect of access control is to preserve the user’s identity accessing a service. This paper’s objective is to propose an access control mechanism of smart contracts to ensure that only authorized users can access the authorized component of the object in the Hyperledger framework.

Smart Contracts, which are embedded in block-chains, allow for the automatic fulfillment of contractual obligations without the need for a reliable third party. Due to this, companies can save administration and service costs, and improve their processes which in turn improve efficiency and reduce risks. EthIR framework is one of the most precise instruments available, with a high success rate. Smart-Contracts need to ensure that they have a minimal number of flaws and vulnerabilities is critical. In this work, we present a Control Flow Graph to apply Cyclomatic Complexity for analyzing smart contracts. Our approach uses EtiIR framework, for creating a CFG from an Etherium Virtual Machine smart contract.

Grey wolf optimization (GWO) is one among the most promising swarm intelligence based nature inspired meta-heuristic algorithm that improves its search process by mimicking the search for prey and attacking strategy of grey wolfs. To further improve its performance, here we have hybridized with Jaya algorithm that improves the exploration capability and hence maintains a trade between exploitation and exploration. An extensive simulation work is carried out to make a comparative analysis of our proposed method with respect to original GWO algorithm and three other meta-heuristic based clustering algorithms such as JAYA, PSO and ALO considering Accuracy, Sensitivity, Specificity and F-score performance measures. The proposed method is used to cluster each dataset taken from UCI machine learning repositories and the experiment is conducted for total 12 datasets separately. The statistical test of the proposed model is conducted by performing Friedman and Nemenyi hypothesis test and Duncan’s multiple test. The obtained results from the statistical test show the superiority of our proposed method with respect to other meta-heuristic based clustering methods.

Heart disease is one of the major cause that leads to a reduce the life span of human beings. Hence it is important to find abnormal heart conditions at an early stage that helps to avoid sudden cardiac death. The optimization algorithms are adaptable and flexible enough to handle difficult non-linear problems. A heart disease prediction model is proposed using Bayesian Optimization of Support Vector Machine, Salp Swarm Optimized Neural Network Classifier, Particle Swarm Optimization, Ant colony, Gradient Descent and Gradient Descent + Particle Swarm Optimization to identify the presence of heart disease. The Proposed GD+PSO mixed optimization algorithms produced the highest accuracy with 99.92% on Cleveland dataset. The proposed algorithms’ superiority is supported by numerous numerical, statistical, graphical, and comparative analyses involving numerous state-of-the-art algorithms. Finally, GD+PSO is suggested in this study for unconstrained optimization problems based on overall performance.

As electricity is a high demand technology in any society and city, the technology of distribution and billing must evolve within a certain time.There are lots of limitations present in current techniques, like we can’t access live information of previous unit usage, users’ privacy is also not maintained here, and one more important thing is that there is no transparency in the payment system. To solve such types of problems, we present a blockchain-based model for usage in smart cities. This will not only maintain transparency but will also maintain the anonymity of users. This paper determines how we can implement blockchain technology in smart cities to facilitate the development of smart cities and delivers a smart city ecosystem based model which depends on smart metres using blockchain technology, which will also build a smart contract between citizens and administrations. This paper will show how the reading of electric metres can be stored in the blockchain and how we can protect the privacy of users using blockchain technology. At the end, citizens will make payments without revealing their privacy. At the end of this paper, we will also conclude how we can do energy management using the data stored in the blockchain.

Modular arithmetic computations are used widely in various data security and reliability techniques. Information security systems certainly benefit from the design of energy-efficient modular exponentiation architectures. The use of low-power logic adders to realize modular exponentiation operations is very essential in prospective cryptography contexts. In this paper, various full adder circuit designs are presented which are used in developing an energy efficient modular exponentiation architecture. Here, the full adder is designed using Register Transfer Level (RTL), Standard Logic Cell (SLC), Reversible Logic Gate (RLG), and Artificial Neural Network (ANN) logic methods. All full adder designs are imposed on modular exponentiation circuit to analyze performance metrics in terms of dynamic power dissipation, Figure of Merit, and Energy Delay Product. The Modular Exponentiation architecture is designed based on the above full adders and is simulated and synthesized using Xilinx Vivado Zynq-7000 family configurable device. From the synthesis results, the dynamic power dissipation, Figure of Merit (FOM), and Energy Delay Product (EDP) of the ANN Modular Exponentiation circuit shows an improvement compared to other designs. The total power consumption, dynamic power dissipation, FOM, and EDP of ANN Full Adder and Modular Exponentiation circuit can achieve (8%, 16%), (23.5%, 20.7%), (14.7%, 14%), and (28.5%, 16%) compared to RLG Full adder and Modular Exponentiation circuit.

Microbial Fuel Cell (MFC) is one of the attractive solution to generate electricity from biodegradable organic matter. But there are different technical challenges such as high source impedance, lower specific power density of MFC and ultra low voltage of MFC which limits the usability of the fuel cell. Different energy harvesting schemes to extract energy from MFC are being investigated by the researchers over the past several years. This work provides a comparative analysis of three distinct power management schemes for MFC.

Underwater Acoustic Sensor Networks (UASNs) is one of the emerging fields in the area of communication due to the number of applications. UASNs face several challenges like limited energy and bandwidth, high bit error rate, packet loss, node mobility, low propagation speed, and routing. Underwater routing is challenging due to the dynamic topology. Many routing protocols used in the UASNs use the hop-count of the neighbor as one of the attributes to select the next-hop. However, due to changes in underwater topology, hop count changes frequently. Obtaining up-to-date hop-count information is one of the major challenges. Many protocols send beacons periodically to update the hop-count, which creates overhead on the network. This paper proposes a Localized Hop-count based Routing (LHR) protocol, which uses a novel mechanism to determine hop-count. Hop-count in LHR is determined based on the local attributes of neighbors. LHR avoids periodic transmission of the beacons from the sink. Thereby, LHR reduces the overhead of transmitting beacons from the sink node periodically. Further, LHR makes use of metrics such as hop-count, depth, and distance for selecting the next-hop.

An underwater sensor network simulator is an analytical tool used to analyze the network performance of a WSN (Wireless Sensor Network). There are various underwater network simulators such as NS2-MIRACLE, SUNSET, Aqua-Net/Mate, DESERT, and UnetStack. However UnetStack is more compatible to real modems from the deployment point of view in comparison with other. UnetStack creates a log-0.txt file and a trace.json file after compiling a groovy file. These trace files are analyzed to get data for per- performance study of a new protocol. To make the process of getting data for performance studies easier, the Underwater Acoustic Sensor Networks Performance Evaluation Tool (UASN-PET) for UnetStack is proposed. This tool helps in extracting and presenting a performance study of a network topology through an interactive GUI. UASN-PET for UnetStack is written in Python so that researchers can spend more time and attention on developing new protocols rather than analyzing trace files. This paper also discusses UnetStack’s trace file format

Various research on digital watermarking has developed great significant progress in recent years but image copyright protection remains a major issue that must be addressed due to the growth and popularization of computer technology. In zero watermarking algorithms, the main tool is continuous orthogonal images for images. Ongoing orthogonal continuous activity is used, consistent in rotation and measurement, and highly improved. Polar Harmonic Fourier Moments (PHFMs) have the ability to define a solid image with excellent performance among the defined continuous orthogonal moments before. PHFM is therefore proposed to deal with images to make a robust image of zero watermarking image suitable for embedding watermark on this paper. Watermarking systems can help with identification and integrity; the importance of the medical image on the other hand should not be altered. Any changes within the cover image may have an impact on the process of decision-making. So, the features of medical images, collected as a set of data will be extracted. We use the Arnold modification of binary sequencing images to be generated for a selected moment to improve security objectives. Then the watermark is embedded in a special function between the scrambled logo image and the binary element of the cover image. A watermarked image abstracted from the PHFMs value without actively embedded to the host image.

One of the most serious issues in today's information systems is data security when communicating over an open network. Digital watermarking is a technology which is rapidly evolving from the recognition of ownership to the recovery of altered information. This work presents dual image watermarking approach formulated on Diffie-Hellman key Exchange (DHKE) algorithm to enhance the secure technique such as: confidentiality, authenticity and non-repudiation. The present dual watermarking technique is intended for safe keeping of copyright by utilizing Integer wavelet transform (IWT) and Singular Value Decomposition (SVD). Before embedding both logos, host images are transformed by IWT. Again, SVD has been applied on selected bands of both the transformed logos. Singular Values (SVs) of both the converted watermark logos are embedded to the sub bands which are transformed by SVD of the original image. The proficiency of present technique has been analyzed in respect of performance parameters such as MSE, SSIM, PSNR, NC. Also, simulation results signify the high embedding capacity of the proposed work.

Federated learning (FL) empowers edge gadgets, similar to the IoT gadgets, servers, and industries that cooperatively prepare a model in the absence of respective privacy information. It expects gadgets that they trade their ML parameters. Subsequently, it expects to be together get familiar with a solid model relies upon the quantity of preparing ventures and boundary time for transmission per each step. Practically, It’s transmissions are regularly overseen at a large number of partaking gadgets on asset-restricted correspondence organizations, for example, remote organizations with restricted data transfer capacity and power. Along these lines, the rehashed FL boundary transmission from edge gadgets instigates a remarkable postponement, which might be bigger than the ML model preparation time by significant degrees. Thus, correspondence delay establishes a genuine issue in FL. A correspondence proficient FL system is proposed to together further develop the FL union time, hence the preparation misfortune. During this system, a probabilistic gadget choice plan is implied such as the gadgets which will altogether further develop the union speed and prepare misfortune to get the high probabilities to choose for the transmission model. We propose a novel technique to additionally decrease the transmission time and to downsize the amount of the model boundaries traded by gadgets, and an effective remote asset designation conspire is created. Reproduction results show that the proposed FL system can further develop identification exactness.

Roadways have always been one of the most used modes of transportation, and their contribution to the nation’s economy is also huge. To meet the demands of the growing global population and an increase in urbanization, there has been an exponential rise in the number of vehicles plying on the roads as well as the length of the roads. With this increase in traffic, coupled with other issues like heavy rainfall, the material used for the construction of the road, etc., the condition of the roads deteriorates with cracks and potholes developing on them, which may lead to serious accidents. For effective maintenance of roads and to reduce the associated risks, these defects must be detected. With the advent of Deep Learning (DL) in the recent past and its applications in various sectors, we have comprehensively explored various approaches, particularly using DL in this study, along with the associated challenges in adopting such techniques and future opportunities in this domain. Based on our analysis, using object detection-based models turned out to outperform other approaches.

Automatic recognition of animal classes by their imageries is an imperative and perplexing task, especially with different animal breeds. Many image classification systems have been projected in the literature but they involve some disadvantages like accuracy deterioration or exhaustive confined calculation. This paper focuses on two methodologies: Transfer Learning and Convolutional Neural Network (CNN) for image-based species identification for distinct animal classes and categorized around twenty-eight thousand animal images from Google Images into ten diversified animal classes. For transfer learning, we have implemented VGG16 (Visual Geometry Group), Efficient NetB2, ResNet101 (Residual Network), Efficient NetB7, and Resnet50 networks that are pre-trained and equated the results of the 5 custom-built CNN networks with these networks using various evaluation metrics that can assist practitioners and research biologists in accurately recognizing various animal species. In terms of performance, VGG-16 attained a maximum accuracy of 0.99 and a Least Validation Cross Entropy Loss of 0.044 for the classification of different species of animals.

There is constant growth in the digitization of information across the world. However, this rapid growth has raised concerns over the security of the information. Today’s internet is made up of nearly half a million different networks. Network intrusions are very common these days which put user information at high risk. An intrusion detection system (IDS) is a software/system to analyze and monitor the data for the detection of intrusions in the host/network. An intrusion Detection System competent in detecting zero-day attacks and network anomalies is highly demanded. Researchers have used different methods to develop robust IDS. However, none of the methods is exceptionally well and meets every requirement of IDS. Machine learning/Deep learning (ML/DL) are among the widely used methods to develop IDS. This proposed technique uses a DL model, Recurrent Neural Network (RNN) with Gated Recurrent Unit (GRU) framework. There are several datasets to evaluate the performance of the learning techniques. CICIDS 2017 is the dataset that contains a variety of cyber-attacks. The proposed technique will use the same to evaluate the deep learning technique. Moreover, the study will also showcase the comparison between the results of existing machine learning algorithms and the proposed algorithm. The comparison would be done on different matrices such as True positive (TP) and False positive (FP) rates, accuracy, precision, etc.

In the 21st-century crop complaint is a serious concern for food security. In this period, utmost of the husbandry support centers and numerous growers use different technologies to ameliorate productivity in farming, but still concerned about plant safety and fast detection of plant leaf diseases which remains difficult in different parts of the regions. Rice plants are frequently infected with diseases that can result in social and financial damage. Many rice crop diseases manifest themselves first on the leaves of the plants. As we say, automated rice plant disease diagnosis is an important aspect of food security, yield loss estimate, and disease management. As a result, computer vision and image processing are utilized to identify infected leave. With the proliferation of digital cameras and ongoing advancements in the computer vision area, automated disease detection techniques are in great demand in precision agriculture, high-yield agriculture, smart greenhouses, and other fields. This study uses an open dataset with 4 types of leaf infections, namely brown spot, blast, bacterial blight, and tungro. In this research, automatically identify plant leaf infection and classify whether the leaf is healthy or diseased, instead of the traditional overlong manual disease diagnostic method, deep CNN models may obtain the best accuracy. We have compared our suggested model (custom-CNN) against pre-trained deep CNN models i.e. VGG19, DensNet121, InceptionV3, as well as ResNet152 deep learning models, and with a learning rate of 0.001, the custom-CNN model obtained greater accuracy of 97.47%. This paper is willing to support and help the farmer Community of the world.

The Voice substitution and audio generation are being used more and more often in a variety of computer listening applications. Furthermore, state-of-the-art perceptual synthesis is allowing richer music without the need for expensive equipment. True audio immersion is when the listener feels what they are listening to, they become part of the story being told. True stereo audio must be generated differently to make use of two channels rather than just one. However, generating stereo audio has not been a popular topic in literature despite being an important component of a listener’s experience. Some great tools for generating stereo audio are Sharp Beta Point or Audacity. This research is focused on developing a generative model for stereo audio generation. It also presents new forms of representation that effectively capture stereo image of stereo audio. It is evident from results that the proposed method improves audio quality to significant degree.

Automated classification of plant leaf diseases is one of the complex concerns in robotics and machine learning fields. Several models have been introduced for detecting and classifying leaf diseases with high classification performance. With the advancement of image processing and machine learning, predicting diseases is the most significant research in recent years. This paper aims to classify lemon leaf diseases by segmenting the diseased part. The lemon leaf dataset is split into training and testing folders. The presented classification method consists of segmentation, feature extraction, and classification steps. For the classification of lemon leaf disease, at first, segmentation is done by using the K-mean clustering algorithm. The infected part of the lemon leaf can be partitioned by the above algorithm. The feature extraction is performed using Gray Level Co-occurrence Matrices (GLCM) method, to extract the texture features from the image. The GLCM method generates the statistical features from the image. The features are passed to the Support Vector Machine (SVM) for the classification of lemon leaf disease. Experiments are conducted for the lemon leaf dataset by taking some samples and the results demonstrate high classification accuracy at a faster speed than other traditional classification methods.

In machine learning, classification is one of the important methods used to train a model for the identification of each class on a labelled dataset. The performance of the algorithm can be enhanced using ensemble techniques such as bagging and random subsampling. Random forest (RF) is an ensemble machine learning algorithm to address multiclass classification problems. Several RF algorithms have been proposed to obtain various levels of accuracy on different datasets during classification. But it has been observed that the manual configuration of RF hyperparameter and its accuracy varies significantly depending upon different datasets. Manual configuration of hyperparameter is a time consuming job. Thus, the hyperparameter optimization process is used to generate optimal hyperparameter efficiently. Hyperparameter optimization also enhances the performance of machine learning algorithms. This paper proposes improved particle swarm optimization (PSO) method as hyperparameter optimizer to identify the optimal hyperparameter of the random forest algorithm. We compare the performance of RF model using improved PSO optimizer and existing RF model with default hyperparameter. The improved PSO outperforms the improvement of accuracy during the classification of each digit of the handwritten dataset.

It is feasible to hypothesize that quantum computers may perform better on certain deep learning applications than classical computers because quantum systems exhibit distinctive patterns that classical systems are assumed not to produce effectively. We explore one such application in this study leveraging Quantum Machine Learning (QML). This study proposes a different approach that provides a comprehensive analysis of telematics data through the extraction of relevant features, followed by feature transformation into a valid weighted risk score. We apply QML frameworks to improve our classification model for the insurance industry by building an experimental Hybrid Classical and Quantum-based Deep Neural Network, a classical Deep Neural Network and a Recurrent Neural Network. Our technique presents an opportunity for insurers to predict a driver’s driving pattern, under different travel conditions, by adding multiple constraints to parameters, for determining personalized premium rates and therefore assign an insurance premium for new drivers or renewal of insurance using predictive modeling. This paper allows insurers to gain a competitive edge by accurate estimation of risk premiums.

Cardiovascular diseases have the highest fatality rate among the world’s most deadly syndromes. They have become stress, age, gender, cholesterol, Body Mass Index, physical inactivity, and an unhealthy diet are all key risk factors for cardiovascular disease. Based on these parameters, researchers have suggested various early diagnosis methods. However, the correctness of the supplied treatments and approaches needs considerable fine-tuning due to the cardiovascular illnesses’ intrinsic criticality and life-threatening hazards. This paper proposes a framework for accurate cardiovascular disorder prediction based on machine learning techniques. To attain the purpose, the method employs an approach called synthetic minority over-sampling (SMOTE). The benchmark datasets are used to validate the framework for achieving better accuracy, such as Recall and Accuracy. Finally, a comparison has been presented with existing state-of-the-art approaches that shows 99.16% accuracy by a collaborative model by logistic regression and KNN.

This study looks at how LSTM networks can be used to predict destiny stock fee patterns primarily based on fee history and technical evaluation. To achieve this, a methodology was developed, several tests were conducted, and the results were evaluated against a number of measures to see if this kind of algorithm outperforms other machine learning techniques. A major current trend in scientific study is machine learning, which involves teaching computers to perform tasks that would require human intelligence. This study uses Long-Short Term Memory Model, to develop a model that predicts future stock market values. The main goal of this paper is to evaluate the predictive accuracy of the machine learning algorithm and the extent to which epochs can improve our model.