Indonesian Journal of Electrical Engineering and Informatics (IJEEI)
Not a member yet
776 research outputs found
Sort by
Leveraging Gradient based Optimization based Unequal Clustering Algorithm for Hotspot Problem in Wireless Sensor Networks
Full text linkWireless sensor networks (WSNs) serve as the basic unit of the Internet of Things (IoT). Because of their low prices and potential use, in recent years, wireless sensor networks (WSNs) have garnered attention for various uses. Then sensor nodes (SN) can prepared with limited battery is critical energy utilization be monitored closely. Hence, reducing the node energy utilization is obviously vital to extending the network lifespan. Clustering is an effectual manner for diminishing energy utilization in WSNs. In a multi-hop clustered network condition, every SN transfers data to its individual cluster head (CH), and the CH gathers the information from its member nodes and relays it to base station (BS) using other CHs. Conversely, the “hotspot” issue is inclined to take place in clustered WSNs while CHs near the BS are heavier intercluster forwarding tasks. In this article, we leverage Gradient based Optimization based Unequal Clustering Algorithm for Hotspot Problem (GBOUCA-HP) technique in the WSN. The GBOUCA-HP technique is applied to get rid of the unequal clustering process in the WSN using metaheuristic algorithms. The GBOUCA-HP technique focuses on the optimization of energy usage, resolving hot spots, and extending the network lifespan. In the GBOUCA-HP technique, the GBO algorithm is based on two concepts such as diversification and intensification search and the gradient‐based Newton’s phenomena. Moreover, the GBOUCA-HP technique adaptive selects the CHs with varying cluster sizes for diverse energy levels and computation abilities of the nodes. The widespread set of simulations and evaluations shows the effective performance of the GBOUCA-HP technique. The GBOUCA-HP technique is found to be a significant approach to tackling the hotspot issue in the WSN with the intention of decreasing energy consumption optimization and enhancing network efficiency
Enhanced Multi-Class Pulmonary Disorder Detection Using Hard Voting Ensemble of CNN Models on X-Ray Images
Full text linkLung diseases represent a major public health concern, requiring timely and accurate diagnosis. Chest X-rays are widely used for initial screening, but manual interpretation is time-consuming and subject to variability among radiologists. To address these challenges, this study presents an automated deep learning-based framework for multi-class lung disease detection. The proposed approach integrates five convolutional neural network (CNN) architectures—EfficientNetB0, DenseNet201, ResNet50, MobileNetV2, and InceptionV3—within a hard-voting ensemble classifier to improve diagnostic performance. Transfer learning is applied to extract deep features from chest X-ray (CXR) images, and the ensemble strategy enhances overall accuracy compared to individual models. The system was evaluated into six categories, including normal, COVID-19, tuberculosis, opacity, bacterial pneumonia, and viral pneumonia. Results demonstrate that the ensemble achieves approximately 97% accuracy, outperforming current state-of-the-art methods. Furthermore, the model shows strong capability in differentiating bacteria from viral pneumonia, underscoring its potential as a reliable tool for automated lung disease diagnosis in clinical practice
AI-Powered CT Scan Enhancement: Turning CTs into MRI Quality Images for Faster and Safer Diagnoses
Full text linkThe use of deep learning (DL) architectures like U-Net and GANs ensures secure, distributed model training across hospitals. The proposed work uses a privacy-preserving federated learning framework for emergency neuroimaging, enabling AI models to convert Computed Therapy (CT) scans into Magnetic Resonance Imaging (MRI) equivalent images as MRI images gives more accurate soft tissue details without compromising patient data. The proposed model integrates DL with saliency maps and Grad-CAM which are the Explainable AI (XAI) tools. The idea is to offer the transparency and build trust in diagnosis of disease. The image quality is measured using the metrics Structural Similarity Index (SSIM) and Paek Signal to Noise Ratio (PSNR) which ensures high-quality image synthesis. The proposed solution enhances the diagnostic accessability in resourse limited hospitals and rural hospitals by preserving patient data with standards. The enhanced model strengthens the framework, privacy techniques and secure aggregation techniques are used to prevent model data leakage during model training or updates. The study provides additional layer of protection to ensures using Federated Learning that even gradient-level information shared between hospitals cannot be traced back to individual patient data. The proposed system is scalable and enables integration with diverse hospital infractures and imaging modalities. The model provides the accessability by turning CT to MRI through secure XAI. The model accuracy ranges to 95% remaining validation accuracy close to train accuracy. The proposed idea provides emergency diagonistics with easy accesibility by preserving privacuy
Partial Discharge Source Identification Using Pulse Height Distribution for Diagnosis in High Voltage Rotating Machines
Full text linkElectrical discharges that are localized in nature and do not completely bridge the electrodes are called as Partial Discharge (PD). PD is the major cause of insulation degradation, and it may eventually lead to system breakdown. Therefore, monitoring of such discharges is important considering efficient diagnosis of high voltage insulation systems. Depending on the source of discharge, there are types of discharges generally occurring in rotating machines viz. Slot Discharge, Delamination, void discharge etc. The plot of count of PD pulses versus the PD magnitude is called the PD height distribution (PDHD) plot or Pulse Height Distribution (PHD) plot. This plot is derived from the actual measurement data on high voltage (HV) rotating machines and the results thus obtained are discussed in this research work. This plot is used to identify the presence of individual PD source or simultaneous occurrence of PD sources in the HV rotating machine. This is the novel contribution of this research work. Phase Resolved PD (PRPD) patterns are used to validate the results. The results for individual discharges and simultaneous discharges are discussed in this paper. The significance of pulse height analysis for diagnosis of HV rotating machines is discussed in this paper
Deep Learning for Arabic Question Classification: Leveraging BERT and Hybrid Neural Networks
Full text linkThis paper presents a deep learning approach for Arabic question classification, leveraging the strengths of pre-trained language models and advanced neural network architectures to address the unique challenges of Arabic text processing. The proposed methodology employs BERT and Word2Vec to generate contextualized and semantic-rich representations of Arabic questions, effectively capturing their linguistic intricacies and morphological complexity. These embeddings are fed into a hybrid classification framework combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks, enabling the extraction of both spatial and sequential features from the input. Experimental results demonstrate the model’s effectiveness, achieving an accuracy of 85.12%, along with high precision, recall, and F1-score metrics. These findings highlight the potential of integrating pre-trained Arabic-specific language models with hybrid deep learning architectures, providing a robust solution for Arabic question classification. This work contributes to advancing Arabic natural language processing, offering a strong foundation for the development of high-performance question-answering systems and related applications
Enhancing Recommendation Systems Through a Hybrid FuzzySparse Similarity Model and Optimizing Noise Reduction
Full text linkThe vast amount of information available on the Internet and e-commerce has led to the development of recommendation systems that help users find relevant products or content. As digital applications continue to grow worldwide, ensuring the right user experience in a short period of time remains a significant challenge. With the increasing use of mobile devices, ensuring access to accurate and timely information has become an essential part of today's business operations. The accuracy of the estimates depends not only on the methodology used but also on the accuracy of the data. External factors and unexpected noise issues can affect users in the rating process. This problematic influence, coming from well-intentioned users, can lead to distortion of the rating results during the recommendation process. In this study, we present a Hybrid Fuzzy-Sparse Similarity (HFSS) methodology designed to improve the accuracy of recommendations and reduce the error caused by source noise due to rating sparsity. Initially, a limited amount of data utilized in a form of rating matrix along with the sparse distribution is collected for the analysis of the recommendation process. An extended fuzzy set matrix creation mechanism is proposed to solve the existing sparsity problems. By using the extended sparse matrix, the similarity of models is calculated from the complex set theory, which allows for model-based recommendations. The proposed HFSS model is evaluated on the MovieLensdatabase. First, the system recommendation performance evaluation is made, and later a comparison performance is measured by MAE, RMSE, and F1 score metrics, which demonstrates better recommendation accuracy and performance than the comparing performance-based methods
Hybrid Stacking Ensemble Model for Breast Cancer Classification: Performance Optimization with Existing Machine Learning Models
Full text linkPrompt and proper recognition of breast cancer Classification (BCC) is imperative in resolution of patient results. In this paper, we suggest a new hybrid Hybrid Stacking Ensemble Model (HSEM), which combines Random Forest, Support Vector Machine and XGBoost classifiers as base learners, and logistic regression as the meta-learner. The HSEM is meant to take advantage of the complement of tree-based and kernel-based algorithms by deriving robust and generalizable binary classification of breast cancer through the use of the Wisconsin Diagnostic Breast Cancer dataset. The accuracy, ROC AUC, and feature importance analysis are considered key metrics that are rigorously tested and compared with traditional standalone models in terms of performance. Findings indicate that the HSEM performs better than traditional classifiers: its accuracy is 99%, and its AUC is 1.00, which makes the method even more viable and reliable when it comes to its prediction values. Learning curves and comparisons further confirm the efficiency of the given approach to be visualized. These results emphasize the possibility of using the Hybrid Stacking Ensemble Model as an efficient instrument of use in medical diagnosis purposes, with the subsequent benefits of providing medical professionals with better diagnostic work support options. The suggested hybrid stacking ensemble not only compares, but also improves performance by combining tree-based and kernel-based learners. The comparative evaluation shows that the optimized hybrid technique always works better than current single-model and ensemble-based methods
Optimizing Energy Management in Hybrid Systems: A Case Study on PV, Battery, and Hydrogen Electrolysis
Full text linkEnergy-intensive facilities face significant challenges in managing energy costs while ensuring reliable power for critical operations. This paper explores the integration of renewable energy and hydrogen technologies in a hybrid system, aimed at reducing grid dependency, minimizing energy costs, and contributing to environmental sustainability. A hybrid energy system comprising solar photovoltaic (PV) generation, battery storage, hydrogen production via electrolysis, and a proton exchange membrane (PEM) fuel cell was developed and simulated using MATLAB Simulink. The system was controlled by an intelligent energy management system (EMS) based on fuzzy logic, which dynamically prioritized energy sources to ensure operational autonomy. A hospital in Jeddah is used as a case study to demonstrate the application of this hybrid system. Simulation results showed that the hybrid system could generate up to 3,017,359 kWh annually, reducing the cost of energy (COE) from 0.11/kWh. The system alleviated grid load by 3,000,000 kWh/year and reduced CO₂ emissions by 1.5 million kg annually. The PV array demonstrated a maximum power point tracking (MPPT) efficiency of 93.6%, and the PEM fuel cell achieved an efficiency of 65%. The fuzzy logic EMS effectively optimized energy flow, ensuring reliable power supply without frequent reliance on grid power. These findings highlight the potential of hybrid renewable energy systems for enhancing energy resilience and sustainability in energy-intensive facilities
Secure Medical Image Transmission via CNN-Derived Keys and Chaotic DNA Encoding
Full text linkThis paper presents a secure and efficient hybrid encryption framework designed for medical image protection. The method combines a fine-tuned MobileNetV3 network for content-adaptive key generation with the nonlinear dynamics of a Lu chaotic system and a DNA-based Cipher Feedback (CFB) diffusion stage. The proposed approach eliminates the arbitrary selection of chaotic maps commonly found in existing methods by dynamically adapting to image content. Experimental tests conducted on brain MRI images demonstrate strong security and robustness, achieving an entropy of 7.9998, NPCR of 99.58%, UACI of 29.93%, PSNR of 7.43 dB, and an average encryption time of 0.24 s. These results confirm excellent randomness, high key sensitivity, and real-time processing capability. The proposed model outperforms recent chaotic and hybrid schemes, making it suitable for secure medical image transmission and telemedicine applications
Trainer Kit for Aroma Classification Using Artificial Intelligence
Full text linkThis research focused on the development and evaluation of machine learning algorithms for aroma classification using sensor data, implemented within the e-Trainose system. Various algorithms, including Neural Network, Support Vector Machines, and Random Forest, were tested to determine their effectiveness in distinguishing between different aroma samples, namely alcohol, coffee, and tea. The study utilized an array of metal oxide semiconductor sensors to capture the volatile organic compounds associated with each aroma. The features tested included sensor responses such as resistance changes and Gaussian smoothing of sensor data. Among the algorithms tested, Neural Network demonstrated the highest accuracy (98.89%), precision (99.10%), recall (99.10%), and F1 score (99.10%), making it the most reliable model for this task. These results highlight the potential of using machine learning with e-Trainose for real-time aroma detection and classification. The research paves the way for future advancements in the field, including the development of hybrid models and further optimization of sensor-based classification systems