International Journal of Communication Networks and Information Security (IJCNIS)
Not a member yet
1021 research outputs found
Sort by
A Constructive Model for Cyber-Attack Prediction Using Efficient Weighted Bi-Directional Learning Approaches
Get PDFAnomaly detection algorithms based on machine and deep learning are currently the most promising techniques for identifying cyber-attacks. However, hostile attacks lower forecast accuracy which is made against these techniques. The resilience of anomaly detection has been measured using a variety of methods in the literature. They neglect to consider the fact that a little disruption in an anomalous sample caused by an assault like a denial of service might cause it to become a genuinely normal sample, but a huge perturbation can transform an anomalous sample into a truly normal sample without affecting the whole system. Even so, it can lead to it being wrongly classified as normal. The approach for determining an anomaly detection model's resilience in industrial contexts is presented in this work. To detect abnormalities brought on by various cyber-attacks; this work used the method of a Support Vector Machine (SVM) for feature extraction and weight analysis. In this case, a unique deep learning-based Bi-LSTM (Bi-directional Long Short Term Memory) only requires a disruption of 60% with 99.6% accuracy of the original sample to create adversarial samples as opposed to the model, which requires a disruption of the entire original sample
Augmenting Cyber Defense Counter To Zero-Day Attacks Through Predictive Analysis- A Fusion Methodology Assimilating Game Theory and RESNet Inspired Optimization Techniques
No full textZero-day attacks pose a significant threat to software vendors, as they exploit previously unknown vulnerabilities, making them insidious and challenging to defend against. Predictive analysis offers a proactive approach to zero-day attack detection, enabling organizations to anticipate and mitigate threats before they manifest. By leveraging advanced techniques such as machine learning and game theory, predictive models can identify emerging attack patterns and adapt in real time to evolving threats. This paper proposes a zero-day attack optimization technique using supervised learning algorithms to identify system disruptions effectively. This paper presents innovative approaches to zero-day attack identification using advanced techniques such as Probabilistic Graph-based Back Propagation Neural Networks, Modified Bi-LSTM with Game Theory, ANN Auto Encoder, Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) and Residual Network (RESNET50). Multiple machine learning approaches were used to determine the most suited model for predicting zero-day assaults. A deep-convolutional n-zero-day network is introduced to distinguish zero-day malware from legitimate software, employing feature selection techniques and a diverse range of machine learning algorithms. This paper presents a novel methodology integrating Hybrid Game Theory (HGT) with Transfer Learning (TL), incorporating feature selection strategies and building upon earlier research. This paper contributes to the field by offering a comprehensive methodology for zero-day attack prediction and highlights areas for further research to address existing limitations and optimize outcomes in network security. Results demonstrate the efficacy of the proposed approach, achieving high detection rates and accuracy in identifying disruptions to network systems
A Novel Feature Reduction Technique for Detection of DoS Attack on Dataset
No full textThe big datasets with numerous features characterizing network traffic parameters are used when creating an Intrusion Detection System (IDS) for machine learning-based IoT attack detection, Thus, one crucial stage in the dataset pre-processing procedure is feature reduction. The current work suggests a filter-based feature reduction technique that aggregates features selected from correlation (CR), relief factor (ReF), and information gain ratio (IGR). Initially, we select a feature subset from the CIE-CICIDS2018 dataset by applying all the above techniques using the average weight threshold. Then further feature optimization is carried out with a subset aggregate strategy (SAS). The proposed SAS feature reduction method with three variants (SAS1, SAS2, and SAS3) efficiently selects common features from three feature subsets formed with three feature reduction techniques. With the SAS strategy, only 57, 27, and 4 features were selected out of a total of 78 features in the CIC-IDS2018 dataset, respectively. Following the feature set reduction, a rule-based PART classifier is used. When metrics like accuracy, precision, recall, and model-building time are compared to the performance of state-of-the-art systems, the recommended method outperforms them
Cloud-Based Secure Blockchain Framework Utilizing Smart Contracts for Regulating Access to Confidential Electronic Medical Health Records
No full textAccess control stands as the single most critical component of cloud storage architecture for enhancing data security. Traditional techniques for data sharing and access control encounter substantial challenges in the research arena, stemming from concerns such as private data leaks and key abuse. With the advent of cloud computing, healthcare applications now have the capability to exchange patient records, thus leveraging the outsourcing of Electronic Health Records (EHR). Moreover, blockchain technology provides a secure foundation by employing multiple encryption methods to verify users. However, patient privacy may be compromised while utilizing the cloud, notwithstanding its benefits in management.This study focuses on two primary outcomes: enhancing access control in blockchain-based cloud systems and refining procedures for exchanging and retrieving data with a focus on prioritizing user privacy. The primary contribution lies in delineating a framework for managing access and data exchange in the cloud using blockchain technology. A well-established method for sharing data and controlling access based on blockchain technology successfully reduces the impact of a cloud infrastructure's single point of failure. Our design integrates off-chain storage solutions to address blockchain's scalability issues while safeguarding EHR privacy and confidentiality. Additionally, we explore the integration of advanced cryptographic techniques to further fortify data security.The proposed system is meticulously crafted to align with existing healthcare standards and facilitate seamless interoperability between different healthcare providers. We evaluate the performance and security of our blockchain-based framework through a series of experiments and compare it with traditional EHR sharing methods. Our findings demonstrate that our framework significantly enhances the security of EHR sharing without compromising the efficiency of data access.
 
Cloud Based DDOS Attack Detection in a Distributed and Collaborative Manner Using Deep Neural Networks
No full textAttacks known as DDOS (Distributed Denial of Service) are now a significant threat to the Internet's availability. Traditional methods of detecting DDOS attacks are limited in their effectiveness due to the heterogeneous nature of cloud data and application deployment. In today’s time, applications are deployed in different containers and nodes over the cloud, and application solutions are deployed in different regions even if they are not fixed with a single cloud service provider. The biggest challenge is data privacy preservation while fighting DDOS attacks while applications are deployed in a distributed manner over the cloud. This study suggests a combination of deep neural networks and federated learning strategies in order to identify distributed denial of service (DDOS) attacks in an environment of heterogeneous cloud service providers. This strategy leverages locally trained models to detect anomalies across different cloud nodes. The models are trained using features extracted from different cloud service providers' logs and traffic data. This strategy is aimed at providing more secure and robust detection of DDOS attacks compared to traditional methods and preserving cloud data privacy. Utilizing several criteria, including accuracy, exactness, recall, and F1-score, the combination of CNN(convolutional neural network) and the federated learning-based detection model is assessed. The outcome of the experiment shows the efficacy of the suggested method in detecting DDOS attacks with high accuracy. This strategy can be used to detect DDOS attacks in an environment of heterogeneous service providers, providing a more secure and robust detection framework
Feature Selection Using a Hybrid Approach: Harmony Search with Recurrent Neural Networks and Filter Methods on the Mimic-Iii Dataset
No full textBuilding reliable and effective prediction models requires careful feature selection, especially when working with complicated medical datasets. Applied to the Medical Information Mart for Intensive Care III (MIMIC-III) dataset, this study presents a novel hybrid method for feature selection that incorporates Harmony Search (HS), Recurrent Neural Networks (RNNs), and conventional Filter Methods. A wide range of clinical records, including vital signs, prescriptions, test results, and diagnostic codes, are included in the dataset. With the help of filter techniques, we guide the first feature ranking in our proposed method (HS-RNN-FM), which combines the deep learning characteristics of RNNs with the global optimization capabilities of Harmony Search. According to experimental data, HS-RNN-FM performs better in terms of precision, recall, accuracy, F1-score, and AUC-ROC than conventional techniques like Genetic Algorithm with Mutual Information (GA-MI) and Particle Swarm Optimization with Chi-Square Test (PSO-Chi2). Due to its exceptional performance, HS-RNN-FM has the potential to improve predictive modeling in healthcare applications by handling huge and complicated datasets in an efficient manner
Disease Detection in Corn or Maize Plant Leaves Using Specim IQ Hyperspectral Imaging and Proposed DNN Classifiers with Alexnet
No full textIt is essential to identify illnesses in corn and maize plants early on in order to preserve crop health and guarantee agricultural output. This work investigates the detection of plant diseases in leaves using advanced deep learning techniques in conjunction with Specim IQ hyperspectral imaging. We compare the performance of a newly constructed classifier, DeepIncepNet, with other state-of-the-art models, such as InceptionV3, ResNet-50, and ResNet-101. We also present a novel Deep Neural Network (DNN) classifier based on the AlexNet architecture. Preprocessing was done on hyperspectral imaging data to improve image quality and retrieve pertinent characteristics.A large dataset was used to train and verify the classifiers, and the results showed excellent disease detection accuracy. The comparative analysis illustrates the benefits and drawbacks of each model, highlighting the possibility for accurate and effective plant disease diagnosis through the combination of deep learning and hyperspectral imaging—a major improvement over conventional techniques
Context-Aware Hate Speech Detection: A Comparative Study of Machine Learning Models
No full textHate speech detection has grown into an essential obligation in today's social media-driven world, where adverse and discriminatory material spreads quickly. This paper conveys an in-depth investigation of recognising hateful speech employing machine learning approaches, with a focus on Logistic Regression and Random Forest models. Sentiment analysis is also investigated as an important factor in differentiating among hateful and non-hateful material, especially on social media, in which jokes and memes can occasionally lack malicious intent. The study contrasts the efficacy of the two models, with a focus on the balance of precision and recall in hate speech recognition. Although the Random Forest model surpassed Logistic Regression, both models struggled to precisely recognise hate speech, particularly in sophisticated and contextualised scenarios. Sentiment analysis is proposed as a future approach for enhancing the identification of potentially hazardous material. The results help to advance the research and development of more precise, context-aware detection of hateful speech systems
The Requirements in Developing Jawi Braille Translator Software
No full textDuxbury Braille Translator (DBT) software comes handy for the persons with visual disability to transcribe reading materials for normal persons and adopt as their own. Even though this software enables transcription of alphabetic texts from various languages into Braille text, it is incapable to directly transcribe the Jawi script into Braille text. Therefore, the study for the development of Jawi Braille Translator software must be carried out to overcome this emerging issue. To achieve that, this article aims to identify the requirements in developing Jawi Braille Translator software. This study adopts qualitative approach, which is by employing semistructured interview method. This study found that the development of Jawi Braille Translator requires elements of JAWS Screen Reader, adding Braille translation for Jawi scripts, sustaining Braille translations for al-Quran and Roman writings, as well as improving technical aspects
The Future of Data Science: Exploring the Convergence of AI, Machine Learning, and Big Data in Transforming Industries
No full textIntegrations of artificial intelligence and machine learning along with big data technologies are impacting industries by improving capacities and performance. This research investigates the impact of these technologies across healthcare, finance, retail, and manufacturing sectors, focusing on the performance of four key algorithms: A Decision Trees, Support Vector Machines (SVM), K-Means Clustering and Neural Networks. The experiments carried out were tested and it was noted that the Neural Networks had a success rate of 92%. 5% and an F1-Score of 0 and an F1-Score of 0 and so on. 89, outperforming other algorithms. SVM had a relatively good success percentage being 88% accurate. as low a figure as 3% with the F1 Score being 0. 85, whereas the performance of Decision trees varies with the accuracy beginning from 75. 0% to 80. 0% and F1-Score from 0 up to 1) and medium (defined as Recall value between 0% and 60% and an F1-Score between 0,61 up to 0,8). 72 and 0. 78. In the experimentation, K-Means Clustering which is useful for the segmentation demonstrated the accuracy of 80. 2% but the training of the model was sensitive to the parameters chosen.] Overall, the results stress the importance of choosing the algorithms according to their capability of meeting the industry requirements and data features. Thus, this study shows that by using big data in combination with AI and ML, it is possible to overcome some of the problems and develop new opportunities for further studies and applications