Evolutionary Computing and Mobile Sustainable Networks

Inhaltsverzeichnis

1. Frontmatter

2. Improved Grey Wolf Optimization-Based Feature Selection and Classification Using CNN for Diabetic Retinopathy Detection

   Anas Bilal, Guangmin Sun, Sarah Mazhar, Azhar Imran

   Abstract

   This research offers a new prediction structure coupling improved grey wolf optimization (IGWO) and convolutional neural network (CNN), called IGWO-CNN, to diagnose diabetic retinopathy. Grey wolf optimizer (GWO) is achieving success among other swarm intelligence procedures due to its broad tuning features, scalability, simplicity, ease of use and, most importantly, its ability to ensure convergence speed by providing suitable exploration and exploitation throughout a search. The suggested methodology used a genetic algorithm (GA) to build diversified initial positions. GWO was subsequently applied to adjust existing population positions in the discrete search procedure, getting the optimal feature subset for a higher CNN-based classification challenge. The presented technique contrasts with GA, GWO and numerous existing state-of-the-art diabetic retinopathy classification approaches. The suggested strategy outperforms all other methods by increasing classification accuracy to 98.33%, indicating its efficacy in detecting the DR. The simulation outcomes have shown that the proposed approach outperforms the other two competing methods.

3. Feature Selection Using Modified Sine Cosine Algorithm with COVID-19 Dataset

   Miodrag Zivkovic, Luka Jovanovic, Milica Ivanovic, Aleksa Krdzic, Nebojsa Bacanin, Ivana Strumberger

   Abstract

   The research proposed in this paper shows application of the sine cosine swarm intelligence algorithm for feature selection problem in the machine learning domain. Feature selection is a process that is responsible for selecting datasets’ features that have the biggest effect on the performances and the accuracy of the system. The feature selection task performs the search for the optimal set of features through a enormous search space, and since the swarm intelligence metaheuristics have already proven their performances and established themselves as good optimizers, their application can drastically enhance the feature selection process. This paper introduces the improved version of the sine cosine algorithm that was utilized to address the feature selection problem. The proposed algorithm was tested on ten standard UCL repository datasets and compared to other modern algorithms that have been validated on the same test instances. Finally, the proposed algorithm was tested against the COVID-19 dataset. The obtained results indicate that the method proposed in this manuscript outperforms other state-of-the-art metaheuristics in terms of features number and classification accuracy.

4. Blood Cell Image Denoising Based on Tunicate Rat Swarm Optimization with Median Filter

   M. Mohana Dhas, N. Suresh Singh

   Abstract

   The significant challenge that occurs due to medical image processing is to acquire the image without the loss of any crucial data. The image data can be degraded by noise or other factors while acquiring or processing the image. This noise affects the image quality since the contrast of medical images is already very low, and it is hard for the experts to identify the infections from the images. Henceforth, image denoising is an essential process in the medical imaging systems. In this paper, a hybrid tunicate rat swarm optimization with median filter (TRSWOAMF) has been proposed to remove or reduce the noise from the medical blood cell image, thus to restore a high-quality image. The proposed TRSWOAMF method uses median filter to remove the noise from the blood cell images, which then optimizes the parameters by using the tunicate rat swarm optimization algorithm. This TRSWOAMF method detects the noise in the blood cell image, wherein the median filter computes the median value for every pixel and best value is replaced. Then the parameters in the image are optimized by the tunicate rat swarm optimization algorithm to retain the original quality of the image. The result shows that the proposed TRSWOAMF method produces high-quality denoised image with a significantly reduced error rate.

5. EVGAN: Optimization of Generative Adversarial Networks Using Wasserstein Distance and Neuroevolution

   Vivek K. Nair, C. Shunmuga Velayutham

   Abstract

   Generative Adversarial Networks (or called GANs) is a generative type of model which can be used to generate new data points from the given initial dataset. In this paper, the training problems of GANs such as ‘vanishing gradient’ and ‘mode collapse’ are reduced using the proposed EVGAN which stands for ‘evolving GAN’. An improvement was done with the help of using Wasserstein loss function instead of the Minimax loss in the traditional GAN. Also, coevolution ensures that the best models from both the generator and discriminator pool are selected for further evolution thereby making the training of GAN more stable. Speciation is also included with a threshold of min. no of species thereby helping in increasing the diversity of the generated image. The model is evaluated in the MNIST dataset and shows better performance in accuracy as well as convergence when compared to the traditional GAN and WGAN.

6. A Hybrid Approach for Deep Noise Suppression Using Deep Neural Networks

   Mohit Bansal, Arnold Sachith A. Hans, Smitha Rao, Vikram Lakkavalli

   Abstract

   Reducing noise to generate a clean speech in stationary and non-stationary noise conditions, or denoising, is one of the challenging tasks in the areas of speech enhancement for single channel data. Traditional methods depend upon first-order statistics, deep learning models, through their power of multiple nonlinear transformations can yield better results compared to traditional approaches for reducing stationary and non-stationary noise in speech. To denoise a speech signal, we propose a deep learning approach called UNet with BiLSTM network (bi directional long short-term memory) to enhance speech. A subset of LibriSpeech speech dataset is used to create training set by using both stationary noise and non-stationary noise with different SNR ratios. The results were evaluated using PESQ (perceptual evaluation of speech quality) and STOI (short-term objective intelligibility) speech evaluation metrics. We show through experiments that the proposed method shows better denoising metrics for both stationary and non-stationary conditions.

7. A New Hybrid Approach of NaFA and PSO for a Spherical Interacting System

   S. Meena, M. Mercy Theresa, A. Jesudoss, M. Nivethitha Devi

   Abstract

   Spherical tanks are widely used as storage system in Petroleum and Chemical industries. The objective of this work is to develop an optimal controller for the liquid level process of the spherical interacting system which is highly non-linear in nature. A hybrid approach called NaFA-PSO has been developed with PSO and Firefly algorithm with neighbourhood attraction model. The developed approach is validated through simulations in MATLAB and it outperformed both the conventional approach (ZN) and the optimization approach, Grey Wolf Optimization (GWO).

8. BitMedi: An Application to Store Medical Records Efficiently and Securely

   Rahul Sunil, Kesia Mary Joies, Abhijeet Cherungottil, T. U. Bharath, Shini Renjith

   Abstract

   Accurate medical history plays a crucial role in the diagnosis by providing physicians relevant data regarding the health of the patient. It helps in preventing prescription errors and the consequent risks to patients. Securing and storing medical records are essential for attaining proper treatment. This paper details the solution BitMedi, a mobile application that helps in uploading medical records and storing them in a well-organized and secure manner. The struggle of all in finding their previous records and risking their lives on giving unsure details of their health to the doctors can hence be eliminated with the solution which helps users save their records in their mobile phones in a user-friendly manner while not compromising on the security of the records. The solution extracts data from the uploaded records while maintaining the user’s privacy and provides graphical representation that helps guide patients to maintain their health.

9. Analysis of Deep Learning Techniques for Handwritten Digit Recognition

   Sagnik Banerjee, Akash Sen, Bibek Das, Sharmistha Khan, Shayan Bhattacharjee, Sarita Nanda

   Abstract

   The huge variations in culture, community and language have paved the path for a massive diversification in the handwriting of humans. Each one of us tends to write in a different pattern. Character or digit recognition finds humongous applications in the recent days especially in the processing of bank statements, sorting of postal mails and many more. Although many classification models exist in literature that successfully classifies the handwritten digits, yet the problem that is still persisting is which one can be termed as an optimal classification model with higher accuracy and lower computational complexity depending upon the circumstances. In this paper, the Machine Learning classification models involving the likes of K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and XGBOOST were compared with that of Deep Learning models like Artificial Neural Network (ANN) and Convolutional Neural Network (CNN). The comparison clearly portrays how convolution operation on images plays a vital role to outperform rest of the classification models. Then the paper compares the CNN models on the basis of two different sets of Loss functions and Optimizers to delineate their role in enhancement of accuracy of the model. The only limitation lies in the fact that in spite of being a handwritten recognizing model this model only recognizes digits in the range of 0–9.

10. Social Media Sentiment Analysis Using the LSTM Model

    A. V. Thalange, S. D. Kondekar, S. M. Phatate, S. S. Lande

    Abstract

    Nowadays, extracting information from social media is providing knowledge about the market and current trends. This paper highlights a novel idea of allowing users to access all such information. A Website is created for visualization of obtained sentiments. It allows users to integrate different social media sites on a single platform for sentimental analysis by providing multiple dashboards under the user’s profile. Users have a choice to provide input from YouTube and/or Twitter to get sentiments. Users need to provide any URL/hashtag. Once input is provided, a trained LSTM model classifies the sentiments as positive, negative, and neutral. Dashboard creation process gets initiated and is deployed on cloud. Classified sentiments are analyzed and are reflected on the dashboard. The dashboard consists of positive, negative, and neutral sentiment’s count, line graph, and pie chart on a real-time basis. These results provide a better understanding of the market and people’s opinion.

11. Developing an Autonomous Framework for Effective Detection of Intrusions

    Sunitha Guruprasad, G. L. Rio D’Souza

    Abstract

    With the growing popularity of Internet, it becomes necessary to protect our systems from imminent security breaches. Apart from the traditional types of attacks, the systems are prone to more sophisticated attacks that originate from various malware systems. Latest developments in the area of intrusion detection systems (IDS) have brought tremendous improvement in detecting the attacks efficiently. But very limited work has been done in the field of autonomous intrusion detection systems. The increased speed and complexity of attacks during recent years show an acute necessity for a more intelligent and autonomous detection mechanism. The proposed IDS involves analyzing the activities of the system and detecting any suspicious behavior on the system. A hierarchical evolutionary model is used to build the autonomous intrusion detection system. A two-phase evolutionary-based method is used in order to detect the intrusions effectively. The first phase generates a list of non-dominated solutions that is used in the second phase for classifying the new packets as normal or intrusive. Results obtained demonstrate very promising results compared to the already existing multi-objective algorithms.

12. Human Health Care Systems Analysis for Cloud Data Structure of Biometric System Using ECG Analysis

    A. Sonya, G. Kavitha, S. Muthusundari

    Abstract

    A cloud security system using cryptography has gained more importance among researchers in recent times. One of the key elements of a cryptography system is that it is capable of converting transactions of the conventional system into an intensive digital transaction module by altering its channels according to the needs of the Clouding or other communication channels. Much research has been conducted with the concept of cryptography, but more importance was given to the medical field, especially for ECG monitoring by a cardiologist which records the electrical signal from the human heart to check for different heart conditions. In this research analysis, the paper tries to analyze the relationship between cryptography and mathematics in the context of the Elliptic Curve (EC). In this paper, ECG encrypted biometric fingerprints have different persons. There are many algorithms in use for obtaining data in the cloud system. The commonly used algorithm for biometric fingerprint encryption is AES, DSA, RSA, and Blowfish. Comparative analysis was carried out in this research with the already developed algorithm and the newly Biometric Cryption of Elliptic Logistic Curve Cryptography [BCELCC] algorithm with the help of Diffihelmen. The system’s computed binary values obtained from the encryption system will be converted into a digital image. The basic patterns of parameters take fingerprint ridges: the arch, loop, whorl, encryption, and decryption time have been given with encryption coding generated first java. The simulation process for the proposed model and the existing model will be carried out in MATLAB coding along with java description. The performance analysis of the outcome of the proposed and existing model will be considered using certain parameters using memory usage, accuracy, comparison, and execution time. The biometric encryption will be carried out for the human being by considering their age, gender, heartbeat, BP level, etc. These data are combined and will be generated as a public key. The main motive of this research is to combine the existing algorithm for biometric encryption along with the proposed algorithm through the security level for storage, and the retrieval process of the cloud of a biometric system will be increased. The security process of the system is considered using low bandwidth and high speed for this research.

13. Sepsis Prognosis: A Machine Learning Model to Foresee and Classify Sepsis

    Vineeta, R. Srividya, Asha S. Manek, Pranay Kumar Mishra, Somasundara Barathi

    Abstract

    Sepsis is caused by bacterial infection that triggers a chain reaction in the body. İt is also called septicemia. If not treated on time, sepsis can lead to tissue damage, organ failure, and death. The aim of present work is to develop a “Machine Learning”-based early warning and “Decision Support System” which can be used to predict whether the person is affected by sepsis or not. The paper enlightens about machine learning module that predicts all the three stages of sepsis namely, sepsis, severe sepsis, and sepsis shock. Experiment is carried out with several classification algorithms to classify the result of person or patient. Performace analysis of algorithms will prove early prediction of sepsis.

14. Feature Engineering of Remote Sensing Satellite Imagery Using Principal Component Analysis for Efficient Crop Yield Prediction

    M. Sarith Divakar, M. Sudheep Elayidom, R. Rajesh

    Abstract

    Early prediction of crop yield before harvest is essential in taking strategic decisions to ensure food availability. Crop yield prediction using remote sensing satellite imagery is a promising approach due to the abundance of freely available data. Machine learning and deep learning techniques used this data to build forecasting systems for mapping crop yield. The high dimension of remote sensing data made training models infeasible using raw pixels. Most of the techniques relied on feature-engineered remote sensing data, whereas recent approaches mainly focused on histogram-based feature engineering. In this study, we followed the histogram-based approach for dimensionality reduction and further reduced the dimension of the input data using principal component analysis. LSTM is used to map crop yield with remote sensing data without losing the temporal properties of the satellite imagery. Results show that the model that used PCA shows comparable performance to existing approaches with fewer parameters.

15. Packet Filtering Mechanism to Defend Against DDoS Attack in Blockchain Network

    N. Sundareswaran, S. Sasirekha

    Abstract

    With the tremendous increase in the Blockchain network scale, and Cryptocurrency network, the Distributed Denial of Service (DDoS) attacks create serious threat for the network operations. The application of Blockchain needs rigorous performance, access requirements, high throughput, and low transmission delay. The proposed method mitigates the DDoS attacks and promised to provide all Service requirements of the Blockchain network. This work mainly puts focus on packet filtering mechanism in real-time to ensure valid users are able to access the service. The flooding of unknown packets are prevented at the source by the novel improved packet marking technique to enable the reasonable service access time to the end user. This work is simulated using NS3 simulator and Ethereum Blockchain platform. It is shown that the traffic is reducing drastically by at least 10% from the actual packets received at the source. It is also proved that the incoming traffic is 90% from the actual traffic flow of packets to the Blockchain network after dropping the unknown flooding of packets by the packet marking technique principle.

16. Data Mining for Solving Medical Diagnostics Problems

    L. A. Lyutikova

    Abstract

    The article solves the problem of creating a software package for computer diagnostics of gastritis. The patient examination indicators and their diagnoses are used as input data. To successfully solve this problem, a logical approach to data analysis is being developed, which allows us to find the patterns necessary for high-quality diagnostics. These laws are identified based on the data provided by specialists and include the results of patient examinations and the existing experience in medical practice in making a diagnosis. Systems of multivalued predicate logic are used for expressive data representation. An algorithm is proposed that implements and simplifies the approaches under consideration. As a result, the developed software package selects the most suitable types of the disease with a predetermined accuracy based on the data of patient diagnostics. If it is not possible to make a diagnosis with a given accuracy based on the results of the examination, then either the accuracy of the decision changes, or it is proposed to undergo an additional examination.

17. Deep Neural Networks-Based Recognition of Betel Plant Diseases by Leaf Image Classification

    Rashidul Hasan Hridoy, Md. Tarek Habib, Md. Sadekur Rahman, Mohammad Shorif Uddin

    Abstract

    Diseases of the betel plant are a hindrance to healthy production which causes severe economic losses and is a major threat to the growing betel leaf industry. This paper introduces an efficient recognition approach based on deep neural networks to diagnose betel plant diseases rapidly to ensure the quality safety and healthy development of the betel leaf industry. A dataset of 10,662 betel leaf images is used for the generalization of deep learning models via the transfer learning technique. EfficientNet B5 has acquired 99.62% training, 98.84% recognition accuracy with 80 epochs. AlexNet, VGG16, ResNet50, Inception V3, and EfficientNet B0 have achieved 81.09%, 83.51%, 86.62%, 94.08%, and 97.29% test accuracy, respectively. During the training phase, AlexNet has taken less time compared to others and misclassified 195 images where EfficientNet B5 misclassified 12 images of the test set. The experimental results validate that the introduced architecture can accurately identify betel plant diseases.

18. Effective Integratıon of Distributed Generation System in Smart Grid

    Namra Joshi, Jaya Sharma

    Abstract

    In this modern era, developing nations are using distributed generation systems’ as a major energy source in deregulated power systems. Distributed generation (DG) is having an important part in improving the quality of human life. This paper emphasis on the various issues concerning the various issues of DG integration into the smart grid and also highlights the benefits and design issues of this integration.