This paper presents characteristics like total dissolved solids (TDS), turbidity, and temperature playing major roles in the health of aquatic organisms; aquaculture productivity mostly depends on preserving ideal water quality. Often insufficient for timely identification of unfavorable situations are conventional manual monitoring techniques. This work demonstrates the evolution of an IoT system based on Arduino with a GSM module for real-time monitoring and alerting. Linked to an Arduino microcontroller for continuous data collection, the system combines a TDS sensor, turbidity sensor, and waterproof temperature sensor (DS18B20). Using algorithms including K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), Logistic Regression (LR), and Random Forest (RF), sensor values are sent via a Wi-Fi module to a cloud server for real-time display and machine learning study. When the system detects unusual water quality circumstances, it immediately sends an SMS alert to farmers running a GSM module (SIM800L/SIM900A), therefore allowing quick corrective action. In areas where aquaculture especially prawn farming is the main way of life, farmers often suffer great losses in a short period of time from unexpected changes in water quality, which can have terrible results, including suicide. By encouraging early intervention, the suggested approach seeks to avoid such results and hence safeguard the farmer’s life as well as the crop. Under both normal and exceptional circumstances, experimental results reveal that the RF algorithm offers the best accuracy and dependability in classifying water quality. By use of predictive analytics and real-time farmer communication, this system presents a low-cost, scalable, intelligent way to improve aquaculture management.
The prediction of cardiovascular disease (CVD) risk requires precise methods which enable doctors to provide treatments to patients who need them most. Conventional statistical and machine learning techniques face difficulties because they cannot understand complex relationships which emerge over time from various patient risk factors. The research presents an advanced deep learning system which combines Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (sGRU) networks with a group-wise feature enhancement mechanism for early CVD detection. The proposed architecture first extracts contextual information in both forward and backward directions using BiLSTM, followed by a GRU layer to create temporal representations which require less processing power. The group-wise enhancement module highlights important clinical feature groups while it reduces the impact of information which has less value. The process uses convolution and pooling layers to extract advanced features while non-essential data is discarded. The experiments are conducted using publicly accessible cardiovascular data which Mendeley and the Cleveland heart disease repository both provided. The proposed model achieved superior performance in comparison to the existing deep learning approaches in terms of accuracy (91.4
Mobile healthcare/medical units (MMUs) enable cost-effective delivery of essential medical services and preventive care to rural and remotely distributed communities where there is a scarcity of qualified personnel and facilities. An important task in planning and deployment of such systems is the provision of efficient routing for the MMUs to follow, such that the number of patients treated (patient coverage) is maximized while adhering to practical budgetary constraints. This problem is formulated in the paper as an extension of the well-known Traveling Salesman Problem (TSP) and is called the Mobile Healthcare Vehicle Traveling Salesman Problem with Constant Budget constraint (MHVTSP-CB). The problem is NP-Hard, due to its extension of the TSP; this motivates the search for heuristic approaches that are able to obtain good solutions to the problem in acceptable time frames. The paper presents novel heuristic and three Hybrid Clustering-based Evolutionary Algorithms (HCEA) for the MHVTSP-CB problem. In order to study their effectiveness, the performance of the proposed hybrid algorithms is compared with that of two well-known metaheuristics - an Artificial Bee Colony (ABC) algorithm and a Particle Swarm Optimization (PSO) approach. All the algorithms are tested on two real-world instances developed as part of this work, derived from the geographic locations of villages of select rural districts in Uttar Pradesh, India. The proposed HCEA3 variant obtains average patient coverage metric values as high as 631,246 and 43,930.57 respectively on the two test benchmarks. This represents a significant improvement vis-a-vis the baseline metaheuristics, which report average coverage values of 108,731.2 and 14,300.0 respectively. The superior results obtained highlight the effectiveness of the proposed methods.
Breast cancer detection using machine learning requires precise understanding of medical images, especially for detecting boundaries between lesions and surrounding tissues. Early detection of cancer is critical since the late detection of the disease could result in fewer options for treatment. Breast ultrasound is one of the imaging methods commonly used for screening and diagnosis because of its widespread availability, lack of ionizing radiation, and capability for real-time imaging. Nevertheless, due to the presence of speckle noise and lack of image contrast, ultrasound images are not easy to interpret, and automated lesion segmentation becomes difficult. In order to overcome the mentioned difficulties, the proposed approach combines deep learning with an advanced image preprocessing for breast ultrasound images. First, the Speckle Reducing Anisotropic Diffusion (SRAD) method is performed to filter out the noise, then the Contrast Limited Adaptive Histogram Equalization (CLAHE) is performed to increase the image contrast. UNet, Attention UNet, Residual UNet, SegNet, and DeepLabV3, were individually trained on the BUSI dataset. Despite achieving notable results in segmentation, these models exhibited few drawbacks such as overfitting to training data, biased predictions, and limited adaptability to diverse patterns within medical images. Hence, to address these issues, we propose the use of ensemble learning. UNet and ResUNet emerged as a powerful combination, attaining a Mean IoU of 0.6762, Mean Dice Score of 0.8070, IoU–Dice Balance of 0.7421, and Accuracy of 0.9634, outperforming the individual models and other ensemble configurations. Further, Explainable Artificial Intelligence (XAI) techniques using Grad-CAM are incorporated to visualize the regions influencing segmentation decisions so as to improve the interpretability and clinical reliability of the proposed framework. The results indicate that combining preprocessing, model comparison, ensemble voting, and interpretability analysis can improve BUSI lesion segmentation performance.
Every individual uses the Internet, making information security a constant concern. The protection of information from a wide range of risks is incorporated into information security. Cybersecurity precisely focuses on safeguarding the data stored online, while information security includes the protection of data across all media. Business owners allow ethical hackers to access their network’s target areas’ vulnerabilities. Ethical hackers deliver recommendations on the best security practices to close these security gaps. Several businesses and organizations use the ethical hacking’s initial phase, known as footprinting method to identify and address network weaknesses and security gaps. Footprinting allows hackers to quickly find and gather target data for breaches. Hackers can use various tools to inform the company about network vulnerabilities. So, the presented survey paper begins with footprinting in ethical hacking, emphasizing the significance of footprinting as an initial stage in ethical hacking, to identify vulnerabilities and gather data about the target system. Then, the integration of Artificial Intelligence (AI) with ethical hacking explores ethical hacking improvement with AI integration for threat detection, automating processes, and precise results by using Machine Learning (ML) and Deep Learning (DL) techniques. Further, several datasets, evaluation metrics, and keywords were used to enhance the knowledge and education in this domain. Moreover, the paper discusses the recent trends, challenges, countermeasures, in ethical hacking, and the use of AI that addresses growing cyber threats. The paper highlights practical applications with case studies that demonstrate the efficacy of AI in ethical hacking in several domains like healthcare, IT, and business. This survey paper provides recent advancements in ethical hacking information under one umbrella.
In recent years, quantum computing overcomes the limitations of traditional computing for solving complex problems which includes image classification tasks. This article aims to introduce a quantum convolution method that can classify brain tumor at a comparable accuracy in less time, with enhanced secure image transfer. The proposed study uses quantum-based deep learning approach QCONV (Quantum Convolution Model) for brain tumor detection. It also ensures secure image transfer between medical entities using cryptographic techniques. The dataset is categorized into meningioma, glioma, and pituitary tumor. The proposed work is divided to two phases where in phase 1 encryption and decryption of images is performed by Advanced Encryption Standard (AES) in Cipher Feedback (CFB) mode to ensure the security and confidentiality of brain tumor MRI images. During decryption AES cipher is then reinitialized with the reconstructed key and IV, allowing the encrypted data to be processed and converted back into its original binary form. In phase 2 the input images are mapped to quantum space using two rotation gates which rotates the qubits over the Bloch sphere and a controlled gate that enables entanglement which further enhances quantum-based classification using QCONV (Quantum Convolution Network). A performance comparison of QCONV with classical deep learning architecture, including DenseNet121, ResNet50, VGG16, and InceptionV3 to evaluate its effectiveness in brain tumor classification. Quantum Convolution model with 4 qubits and quantum depth of 5 has shown accuracy of 97 O(n^2) operations whereas the quantum models exhibit O(log(n)) . The comparative study refers to the asymptotic scaling of the quantum circuit depth and the number of qubits under an amplitude-style state-encoding assumption, and not to the measured regular execution time. The results demonstrate that the proposed QCONV outperforms the existing classical deep learning models in terms of accuracy, sensitivity, specificity, and F1 score.
With the increasing deployment of mobile health services, it is important that the issues surrounding the quality of service and scarcity of resources be considered. The increased prevalence of the Internet of Things (IoT) has created the need for effective wireless communication systems. In particular, the Internet of Medical Things (IoMT) combines mobile and IoT technologies to diagnose, monitor, and provide health services. Applications running on mobile devices are often constrained by Quality of Service (QoS) requirements, making performance difficult due to limited resources. This study introduces a Neural Network (NN)-enhanced evolutionary optimization framework to improve the performance of IoMT systems through energy-efficient task distribution. We combine the Lemurs Optimizer (LO) algorithm with an error-backpropagation training procedure to mitigate NN drawbacks during training and enhance load-offloading efficiency. The MATLAB modeling verifies that the proposed method is superior in energy utilization, cost, and task execution time compared to other options. The findings are beneficial for extending this study and provide informative insights to optimize resource usage in IoMT systems.
This study introduces a novel time-domain diffraction model to analyze Ultra-Wideband (UWB) signal propagation affected by human body obstructions, addressing the critical need for accurate channel modeling in ultra-dense networks (UDNs). Unlike conventional frequency-domain approaches, our innovative time-domain framework leverages single, double, and triple knife-edge diffraction models, uniquely integrating apical diffraction to capture complex signal interactions around the human body. We propose a computationally efficient method to model dynamic attenuation variations based on the body’s relative position, offering superior precision over existing models. Validated through extensive simulations across sub-6 GHz, mmWave, and UWB bands, our results show the triple knife-edge model outperforms state-of-the-art methods with up to 15
Photovoltaic (PV) Systems that connect to the electrical grid are now a primary vector for transitioning from conventional to renewable energy by integrating scalable and clean electricity into the existing electrical grid infrastructure. However, intermittent generation of PV electricity combined with the effects of the environment, including weather, and the connection of power electronics, can introduce power quality problems such as voltage fluctuation and harmonic distortion at the point of common coupling (PCC). This study provides a comparison of Power Quality (PQ) improvement via Total Harmonic Distortion (THD) analysis in a conventional (AC) 0.3 MW Grid Connected PV System with Incremental Conductance (IC) Maximum Power Point Tracking (MPPT) Techniques under moderate to high insolation and temperature conditions, simulated within the PSCAD software environment. The IC simulations at 500 W/m2, 25 °C and 1000 W/m2, 35.5 °C produced stable PV Voltage and Current profiles, produced balanced three-phase Inverter output, and produced voltage THD values that were below IEEE 519 and CEA limits consistently < 2.18
The advent of synthetic voice generation technology, it has created transformative capabilities and brought about significant challenges, especially in voice authentication, digital forensics, and security. Deepfake audio is one type of deepfake content that is artificially created or altered to sound like a human voice and can be highly dangerous through identity theft, misinformation, and exposure of confidentiality. True identification of such digitally imitated voices is critical in ensuring that we are curtailing these risks and ensuring the sanctity of the voice-based systems. The study presents a systematic process of Digitally Generated Voice Detection using Machine Learning and Deep Learning systems. The data were properly prepared in a manner that separates the real and fake audio files, representing the extracted features in isolation and using audio files to train Random Forest, Deep Neural Network (DNN) model, XGBoost, and a combination of DNN + XGBoost model. Characteristics such as HNR, pitch variance, frequency range, intensity, mean pitch, chroma, mel spectrogram, and spectral contrast are extracted from the signals. MFCCs, chroma, mel, and spectral contrast are relatively common in speech and audio processing because they represent the important features of an audio signal, including how the human auditory system responds to sound. This extraction process gives the system the capability to notice minute variations between human and artificial voices, thus playing an essential role in the detection of the voice characteristics. These results indicate that the combination of DNN and XGBoost model could perform the best in terms of accuracy, in terms of precision, and computation. This research addresses the increasing threats that deepfake audio represents, thus advancing the security frameworks, preventing the potential harm that this type of voice fraud can cause, and developing the acoustic analysis of sound. The proposed system will offer an easy-to-scale, advanced technology to differentiate between the human and artificial voice and potentially form the basis of more advanced uses of deepfakes detection.
Accurate rainfall prediction is critical for ensuring agricultural resilience, disaster preparedness, and energy resource planning, particularly in monsoon-driven regions like West Bengal, India. This study introduces an efficient machine learning framework that leverages ensemble techniques for high-accuracy annual precipitation forecasting. The methodology utilizes historical climate data from NASA’s Power Data Access Viewer (1984–2022), incorporating features such as temperature, humidity, wind speed, and previous rainfall. An extensive data preprocessing pipeline was employed, including label encoding, outlier detection, normalization, dimensionality reduction using Principal Component Analysis (PCA), and class balancing via SMOTE and SMOTE-Tomek. Feature selection was performed using correlation analysis, chi-square testing, wrapper methods, and embedded models. Multiple ensemble approaches, bagging, boosting, and stacking, were benchmarked using tenfold cross-validation. The highest accuracy of 99.38
There are many challenges for Autonomous Underwater Vehicles (AUVs) in the ocean because hydrodynamic coefficients, payload alterations and disturbances from outside can all seriously affect their stability and ability to function. It is very important to maintain stability and good performance for successful underwater work. Details of a robust control method using μ-synthesis are described in this paper to solve the problems caused by uncertainties in AUVs. It assumes all uncertain parameters are included inside structured bounds and makes sure stability and effectiveness are maintained by applying robust control methods across many working conditions. Tests using simulation confirmed that μ-synthesis improves both the robustness and maneuvering skills of AUVs when there are a lot of parametric changes. The study proves that μ-synthesis can successfully control AUVs in uncertain conditions, as it improves accuracy, lessens disturbance effects and boosts stability margins.
5G NR utilizes both standalone architecture (SA) and non-standalone (NSA) architecture to provide seamless connectivity to the end users. The difference between SA and NSA lies in the core network in which NSA utilizes the 4G core network services whereas the SA utilizes 5G core network services. These two architectures utilize different frequency bands in multi radio access network (RAN) scenarios such as n78 and carrier aggregation with n28 band under frequency range 1 (FR1), and n258 band under frequency range 2 (FR2). To fulfill the requirements of the end users, the selection of an appropriate frequency band while deploying 5G RAN services is essential. In this paper, the performance analysis of three frequency bands viz. n78, n28, and n258 is carried out by considering three gNBs. The performance analysis is carried out in terms of throughput, packet drop ratio, and total application delay. The simulation consists of different traffic types (i.e., constant bit rate (CBR), voice, and video services) with different QoS classes on the 5G NR QoS framework. In addition, each service consists of separate QoS flows and priority for scheduling of the resources. Experimental evaluations indicate that the carrier aggregation technique effectively reduces the hardware cost and provides almost equal throughput among all other gNBs. Here resource scheduling is found to be an important key paradigm that can perform mapping as per QoS requirements specified in 3GPP specifications.
The emergence of IoT and its applications have enforced different security challenges to identify unauthorized users. Authenticator is one of the applications which is used to provide multi fold security for better robustness. Still there is a possibility that some unauthorized users will try to access the applications. In this article, we present a comprehensive exploration of user-centric analysis and suspicious user detection, specifically focused on the authentication process within the Authenticator application. With cybersecurity being of paramount importance, the study employs advanced machine learning techniques to analyze user interactions and activities, aiming to identify and flag potentially suspicious behavior within individual user accounts. The Authenticator multi-factor authentication system, encompassing email-password, One-Time Password (OTP), and push notification steps, forms the basis for analysis. The study’s motivation lies in safeguarding user accounts from unauthorized access and fraud, necessitating proactive measures against evolving cyber threats. The approach involves processing unstructured, unsupervised data from Elasticsearch and Kafka, extracting valuable insights through feature aggregation, temporal analysis, and geospatial aspects. Evaluation employs the Silhouette Score to measure k-means clustering quality, as well as in the Isolation Forest model, contributing to effective suspicious user detection. During the prediction phase, we retrieve a master dataframe from the SQL database, which contains patterns of both suspicious and normal user behaviors. Utilizing the k-nearest neighbors (KNN) algorithm, we identify the nearest matching pattern from this master dataframe and assign that label to our test data. The study’s outcomes enhance security in the Authenticator application by distinguishing normal and suspicious login patterns, strengthening the multi-factor authentication process for increased reliability.
Healthcare analytics aims to derive profound insights and make accurate predictions, which makes preserving critical patients’ data privacy. To solve this problem, this research implements a Federated Learning Framework to execute Whale Optimization for Health Privacy Preservation Analytics (WHOPPA) of healthcare records. Employing the Whale Optimization Algorithm, a metaheuristic optimization approach is modelled with the social behaviour of humpback whales. The primary purpose of this methodology is to strive for optimum performance while retaining privacy through federated learning in various healthcare sectors. The WHOPPA framework will be constructed to adapt to varying data and ensuring privacy requirements by healthcare organization with advanced analytics tools. Proposed WHOPPA is benchmarked against PSO-based, Genetic Algorithm-based, and Ant Colony Optimization-based models. The WHOPPA system demonstrates efficiency in privacy preservation and optimization of datasets, attaining 87
Traffic Sign Recognition (TSR) is one of the crucial steps for enabling self-driving vehicles to analyze the sign boards and take necessary decisions to enhance road safety and navigation. This paper implements multiple deep learning (DL) architectures and optimizes their hyperparameters using swarm intelligence, namely Firefly optimization. To improve recognition performance, images undergo initial pre-processing, including resizing and grayscale conversion. These refined images are then represented as matrix structures to serve as inputs for various deep learning architectures. The proposed framework is tested on two datasets: i.) Chinese Traffic Sign Dataset (CHSTD) and ii.) German Traffic Sign Recognition Benchmark (GTSRB). The performance analysis on benchmark datasets indicates that the CNN outperforms other DL architectures by achieving an average accuracy of 96.67
Most of the applications in real world is connected through internet or intranet through a well-supported network layer. The data captured through physical layer traverses through nodes in network layer. During this process the QoS must be maintained effectively for optimized point to point or multipoint communication. The nodes are connected, in order to maintain the connectivity and chose an efficient path nodes are clustered and Cluster Head are assigned to monitor the QoS of the network layer. In this research work the Static type of network is considered, its quality is enhanced in terms of attributes such as energy efficiency, network lifetime, etc., the nodes are used for routing packets, the clusters are formed in circular regions and transmission and reception paths are established. In this work C-LEACH which is the standard WSN algorithm is improved by adapting Circular Clustering (CC) before forming a Cluster Head, the algorithm improves the QoS of the WSN.
The Indian Railway system is the fourth largest rail network in the world by size, carrying approximately 24 million passengers daily. In such a huge, dynamic, real-time system, knowing the inconveniences of individuals and optimally taking actions is beyond the scope of the human-operable system. For this, an automated intelligent system capable of recognizing human sentiment from natural language is required. This challenge is heightened because India is a multilingual country. Though some previous studies tried to address this problem by considering Hindi and English, to the best of our knowledge, none have considered Bengali. Bengali is the second most popular language in India, with around 90 million native speakers, and neglecting the mood of the Bengali people will make any system incomplete. To address this gap, a bilingual, automated, natural-language-based mood-detection model was proposed that incorporated Bengali (written in Bengali font), English (written in Latin font), and Bengali (written in Latin font). We created a custom dataset for this purpose by collecting feedback from multiple social platforms. Sentiment detection was achieved through the assignment of weighted values to multilingual words and emojis, and then the model was trained. Internally, various classification algorithms were used to evaluate the effectiveness of the proposed system. Specifically, Logistic Regression, Support Vector Machine, K-Nearest Neighbor, and Decision Tree classifiers were employed for evaluation. Upon applying these algorithms, performance metrics such as accuracy, precision, recall, F1-score, and cross-validation score were calculated to measure their effectiveness comprehensively. The proposed model achieved a 93
In today’s fast moving era, specially in urban world, crowd anomaly detection and continuous crowd behaviour monitoring is necessary to ensure public safety. This study reviews and give comparison of deep learning techniques such as “convolutional neural network” and “auto encoder” based models which is used in crowd anomaly detection and analysis. CNN based frameworks like YOLO and hybrid CNN LSTM models works well in extraction of spatial features and enables high accuracy detection in real time. While, autoencoder architectures contains spatio temporal and memory augmented variants and learn common behavioural patterns without labelled anomalies. Comparative evaluations shows that CNNs gives fast inference and accurate localization while autoencoders perform best in identifying new or unknown events and handling the shortage of labelled datasets. Overall, Deep learning based crowd anomaly detection shows significant progress toward automated, smart and intelligent monitoring. However, some challenges like occlusion, domain shifts and privacy concerns will need to be solved. The study concludes that by integrating lightweight frameworks with multi modal sensing and ethical design can make robust, adaptive and privacy preserving crowd management systems. This brings the research closer to the real world deployment and secure public environment.
Landslide detection is one of the key challenging tasks for disaster risk reduction and developing early warning systems. The conventional deep learning models such as U-Net and YOLO often fail to capture fine-scale features in heterogeneous satellite imagery. To overcome these limitations, we propose a Fusion-Aware Unified Framework that integrates deep fusion techniques such as Early Fusion and Late Fusion with advanced deep learning models such as attention mechanisms. Two datasets were used throughout this study, namely Landslide4Sense and HR-GLDD. We began by implementing basic U-Net and YOLO integrated with deep fusion techniques independently to identify landslides detection limitations. This approach established a baseline performance for comparison with more advanced models. Based on these results, two attention-driven fusion models were developed namely Attention-Driven Early Fusion model and Attention-Driven Late Fusion model. Following the evaluation of these models, we further extended the work by incorporating a self-attention mechanism with fusion techniques. We proposed two self-attention-based architectures named as SAMEL (Self-Attention Mechanism with Early Fusion), and SAMSNet (Self-Attention Mechanism with Stacking Network – A Late Fusion Technique) respectively. These models uses a self-attention module to effectively capture long-range spatial dependencies for landslide delineation. An ablation study was performed on different loss functions to evaluate their impact on models’ performance. Experimental results demonstrates that self-attention-guided models are effective in capturing accurate landslides. SAMEL achieved an accuracy of 98.47
