Due to the monumental growth of Internet applications in the last decade, the need for security of information network has increased manifolds. As a primary defense of network infrastructure, an intrusion detection system is expected to adapt to dynamically changing threat landscape. Many supervised and unsupervised techniques have been devised by researchers from the discipline of machine learning and data mining to achieve reliable detection of anomalies. Deep learning is an area of machine learning which applies neuron-like structure for learning tasks. Deep learning has profoundly changed the way we approach learning tasks by delivering monumental progress in different disciplines like speech processing, computer vision, and natural language processing to name a few. It is only relevant that this new technology must be investigated for information security applications. The aim of this paper is to investigate the suitability of deep learning approaches for anomaly-based intrusion detection system. For this research, we developed anomaly detection models based on different deep neural network structures, including convolutional neural networks, autoencoders, and recurrent neural networks. These deep models were trained on NSLKDD training data set and evaluated on both test data sets provided by NSLKDD, namely NSLKDDTest+ and NSLKDDTest21. All experiments in this paper are performed by authors on a GPU-based test bed. Conventional machine learning-based intrusion detection models were implemented using well-known classification techniques, including extreme learning machine, nearest neighbor, decision-tree, random-forest, support vector machine, naive-bays, and quadratic discriminant analysis. Both deep and conventional machine learning models were evaluated using well-known classification metrics, including receiver operating characteristics, area under curve, precision-recall curve, mean average precision and accuracy of classification. Experimental results of deep IDS models showed promising results for real-world application in anomaly detection systems.

The potato plant is amongst the most significant vegetable crops farmed worldwide. The output of potato crop production is significantly reduced by various leaf diseases, which poses a danger to the world’s agricultural production in terms of both volume and quality. The two most destructive foliar infections for potato plants are early and late blight triggered by Alternaria solani and Phytophthora infestans. In actuality, farm owners predict these problems by focusing primarily on the alteration in the color of the potato leaves, which is typically problematic owing to uncertainty and significant time commitment. In these circumstances, it is vital to develop computer-aided techniques that automatically identify these disorders quickly and reliably, even in their early stages. This paper aims to provide an effective solution to recognize the various types of potato diseases by presenting a deep learning (DL) approach called EfficientPNet. More specifically, we introduce an end-to-end training-oriented approach by using the EfficientNet-V2 network to recognize various potato leaf disorders. A spatial-channel attention method is introduced to concentrate on the damaged areas and enhance the approach’s recognition ability to effectively identify numerous infections. To address the problem of class-imbalanced samples and to improve network generalization ability, the EANet model is tuned using transfer learning, and dense layers are added at the end of the model structure to enhance the feature selection power of the model. The model is tested on an open and challenging dataset called PlantVillage, containing images taken in diverse and complicated background conditions, including various lightning conditions and the different color changes in leaves. The model obtains an accuracy of 98.12% on the task of classifying various potato plant leaf diseases such as late blight, early blight, and healthy leaves in 10,800 images. We have confirmed through the performed experiments that our approach is effective for potato plant leaf disease classification and can robustly tackle distorted samples. Hence, farmers can save money and harvest by using the EfficientPNet tool.

A vehicular ad hoc network (VANET) is a sophisticated wireless communication infrastructure incorporating centralized and decentralized control mechanisms, orchestrating seamless data exchange among vehicles. This intricate communication system relies on the advanced capabilities of 5G connectivity, employing specialized topological arrangements to enhance data packet transmission. These vehicles communicate amongst themselves and establish connections with roadside units (RSUs). In the dynamic landscape of vehicular communication, disruptions, especially in scenarios involving high-speed vehicles, pose challenges. A notable concern is the emergence of black hole attacks, where a vehicle acts maliciously, obstructing the forwarding of data packets to subsequent vehicles, thereby compromising the secure dissemination of content within the VANET. We present an intelligent cluster-based routing protocol to mitigate these challenges in VANET routing. The system operates through two pivotal phases: first, utilizing an artificial neural network (ANN) model to detect malicious nodes, and second, establishing clusters via enhanced clustering algorithms with appointed cluster heads (CH) for each cluster. Subsequently, an optimal path for data transmission is predicted, aiming to minimize packet transmission delays. Our approach integrates a modified ad hoc on-demand distance vector (AODV) protocol for on-demand route discovery and optimal path selection, enhancing request and reply (RREQ and RREP) protocols. Evaluation of routing performance involves the BHT dataset, leveraging the ANN classifier to compute accuracy, precision, recall, F1 score, and loss. The NS-2.33 simulator facilitates the assessment of end-to-end delay, network throughput, and hop count during the path prediction phase. Remarkably, our methodology achieves 98.97% accuracy in detecting black hole attacks through the ANN classification model, outperforming existing techniques across various network routing parameters.

Restricted abilities of mobile devices in terms of storage, computation, time, energy supply, and transmission causes issues related to energy optimization and time management while processing tasks on mobile phones. This issue pertains to multifarious mobile device-related dimensions, including mobile cloud computing, fog computing, and edge computing. On the contrary, mobile devices’ dearth of storage and processing power originates several issues for optimal energy and time management. These problems intensify the process of task retaining and offloading on mobile devices. This paper presents a novel task scheduling algorithm that addresses energy consumption and time execution by proposing an energy-efficient dynamic decision-based method. The proposed model quickly adapts to the cloud computing tasks and energy and time computation of mobile devices. Furthermore, we present a novel task scheduling server that performs the offloading computation process on the cloud, enhancing the mobile device’s decision-making ability and computational performance during task offloading. The process of task scheduling harnesses the proposed empirical algorithm. The outcomes of this study enable effective task scheduling wherein energy consumption and task scheduling reduces significantly.

Brain Tumor is a significant health challenge, with early, precise detection crucial for improving patient outcomes. Traditional diagnostics methods depend on expert radiologists, leading to subjectivity and time-intensive evaluations. In recent advancements, AI and DL have demonstrated remarkable medical imaging potential, yet many existing models are limited to binary classification, failing to distinguish between multiple tumor types. This limitation restricts their clinical applicability and effectiveness in comprehensive diagnosis. We propose a deep learning-based multi-class classification approach for brain tumor detection, leveraging EfficientNetV2 and Vision Transformer (ViT) models to address this challenge. EfficientNetV2, known for its optimized convolutional architecture, achieves high accuracy while maintaining computational efficiency, whereas ViT, a transformer-based model, effectively captures global contextual information in medical images. Our approach was evaluated on a brain tumor MRI dataset containing 7,023 images, divided into training and testing sets. EfficientNetV2 achieved an accuracy of 95% with a loss of 0.13, alongside an F1-score, precision, and recall of 0.96. In comparison, ViT attained 90% accuracy with a loss of 0.30 and an F1-score, precision, and recall of 0.89. The highest accuracy 96%, was achieved by the proposed model using the geometric mean ensemble learning technique. The results demonstrate that our proposed system outperforms many existing deep learning models, offering a robust solution for multi-class brain tumor classification. Our approach enhances diagnostic accuracy by integrating convolutional and transformer-based architectures, aiding radiologists in early and automated tumor detection. This research contributes to the advancement of AI-driven medical diagnostics and has the potential to improve clinical decision-making in brain tumor assessment.