Articles

Access the latest knowledge in applied science, electrical engineering, computer science and information technology, education, and health.

Advanced Search
Filter Icon

Filters article

Years

FAQ Arrow
Single Year
Range Year
0
0

Source Title

FAQ Arrow

Authors

FAQ Arrow

31,456 Article Results

Anomaly-based intrusion detection leveraging optimized firewall log analysis: a real-time machine learning solution

10.11591/ijece.v15i5.pp4785-4802
Tran Cong Hung , Dam Minh Linh , Han Minh Chau , Ngo Xuan Thoai , Thai Duc Phuong , Huynh De Thu
Firewall logs play a vital role in cybersecurity by recording network traffic and flagging potential threats. This study evaluates five machine learning algorithms-decision tree (DT), random forest (RF), extra trees (ET), CatBoost (CB), and AdaBoost (AB)-on a dataset of 65,532 firewall log entries. Models were assessed using accuracy, precision, recall, training/prediction time, and Pearson correlation for feature selection, across multiple train-test splits. The DT model achieved the best performance, reaching 99.45% test accuracy, 97.457% precision, and 93.389% recall at a 7:3 split, along with the fastest training time (0.20642s). We propose real-time flow-level intrusion detection (RT-FLID), novel, lightweight, real-time intrusion detection system that leverages multithreaded processing and flow-level analysis to boost detection speed and scalability. Unlike existing approaches that rely heavily on deep packet inspection or computationally intensive processing, RT-FLID requires minimal resources while maintaining high detection accuracy. The architecture efficiently handles large traffic volumes and dynamically identifies anomalies such as distributed denial-of-service (DDoS) and port scans. Validated on real-world logs, the system maintained high accuracy in critical classes like “deny” and “reset-both.” These findings highlight RT-FLID’s novelty and practical advantages, demonstrating its potential for deployment in high-throughput, low-latency network environments.
Volume: 15
Issue: 5
Page: 4785-4802
Publish at: 2025-10-01

Enhanced solar panels fault detection approach using lightweight YOLO

10.11591/ijai.v14.i5.pp3554-3562
Naima El Yanboiy , Mohamed Khala , Ismail Elabbassi , Nourddine Elhajrat , Omar Eloutassi , Youssef El Hassouani , Choukri Messaoudi
Artificial intelligence (AI)-driven fault detection improves the reliability of solar energy by reducing the chances of system failures. However, existing single-stage object detection methods excel in accuracy but demand high computational resources, preventing seamless integration into embedded systems. This paper introduces a lightweight approach using YOLOv5, which incorporates a multi-backbone design, specifically tailored for accurate fault detection in solar cells. It evaluates YOLOv5 and TinyYOLOv5. The findings emphasize the effectiveness of YOLOv5l with Ghost backbone, particularly notable for its precision rates of 96% for faulty and 93% for non-faulty instances. Additionally, it showcases commendable mean average precision (mAP) scores, achieving 78% at an intersection over union (IoU) threshold of 0.5 and 72% at an IoU of 0.95. Additionally, YOLOv5_Ghost emerges as the optimal selection, showcasing competitive precision, processing speed of 42.1 giga floating point operations per second (GFLOPS), and remarkable efficiency with 2.4 million parameters. This evaluation underscores the effectiveness of YOLOv5 models, thereby leading to advanced solar energy technology significantly.
Volume: 14
Issue: 5
Page: 3554-3562
Publish at: 2025-10-01

Enhanced classification of aromatic herbs using EfficientNet and transfer learning

10.11591/ijai.v14.i5.pp4123-4136
Samira Nascimento Antunes , Madalena De Oliveira Barbosa Divino , Luana Dos Santos Cordeiro , Fernanda Pereira Leite Aguiar , Marcelo Tsuguio Okano
Herbs have long been used for culinary and medicinal purposes, as well as in religious rituals, due to their essential oils and aromatic properties. However, distinguishing between aromatic and medicinal herbs based on visual characteristics alone can be challenging. With recent advances in computer vision, plant identification from images has seen significant growth, offering promising applications in several domains. This article aims to evaluate the classification of aromatic herbs using the EfficientNet convolutional neural network (CNN) technique with transfer learning. The methodology used is experimental research, systematically manipulating variables to observe their effects on the object of study. The researcher plays an active role in this process, rather than being a passive observer. Based on the results and the literature review, it is evident that the objective of this research was achieved, as despite the opportunities for improvement in training to achieve accuracy above 0.8, it was possible to evaluate the classification of aromatic herbs using EfficientNet CNN through the transfer learning technique.
Volume: 14
Issue: 5
Page: 4123-4136
Publish at: 2025-10-01

Stock market liquidity: hybrid deep learning approaches for prediction

10.11591/ijai.v14.i5.pp3624-3633
Mariam Ait Al , Said Achchab , Younes Lahrichi
Predicting stock market liquidity especially in emerging or frontier financial markets, such as the Casablanca stock exchange (CSE), presents significant challenges given the relative narrowness and volatility of these markets. In this paper, we conduct a comprehensive study to address the predictions accuracy gaps between five main deep learning models: convolutional neural network (CNN), long short-term memory (LSTM), bidirectional LSTM (BiLSTM), and two hybrid architectures, CNN-LSTM and CNN-BiLSTM. The proposed methodology focused on training and testing these models on historical data from the CSE, with precision on capturing both spatial and temporal market dynamics. The models were fine-tuned using key hyperparameters and validated on 20% of the dataset to ensure reliable results. The evaluation of performance was conducted using error metrics such as mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE). The study demonstrates that the hybrid CNN-biLSTM model consistently outperformed all standalone and other hybrid models in predictive accuracy. This underscores the considerable promise of hybrid deep learning architectures for addressing the unique challenges of predicting stock market liquidity in volatile and emerging financial markets.
Volume: 14
Issue: 5
Page: 3624-3633
Publish at: 2025-10-01

Enhancing concrete sustainability: a neural networks hybrid optimization approach to predicting compressive strength using supplementary cementitious materials

10.11591/ijece.v15i5.pp4965-4982
Esra’a Alhenawi , Ayat Mahmoud Al-Hinawi , Zaher Salah , Omar Alidmat , Esraa Abu Elsoud , Raed Alazaidah , Bashar Rizik AlSayyed
This research evaluates the implementation of advanced machine learning methodologies for concrete mix design to achieve better predictive models and sustainable outcomes. This study develops a hybrid optimization approach by combining dung beetle optimizer (DBOA) and firefly algorithm (FLA) to optimize hyperparameters for convolutional-recurrent neural networks in order to correctly predict concrete compressive strength when using supplementary cementitious materials (SCMs). Shapley additive explanations (SHAP) provide feature significance analysis, which ensures that the model produces understandable conclusions supported by empirical findings. The findings demonstrate that this method enhances the predictive accuracy of strength analysis, along with offering critical insights about SCM usage in order to improve sustainable construction methods. The model proves suitable for integration into actual concrete mix design and quality control systems because it achieves both computational speed and passes validation tests on distinct datasets. The research creates foundations for upcoming studies about multimodal learning enrichment and deals with ethical concerns in construction site safety when using machine learning systems.
Volume: 15
Issue: 5
Page: 4965-4982
Publish at: 2025-10-01

Optimized convolution neural network with ant colony algorithm for accurate plant disease detection

10.11591/ijai.v14.i5.pp3724-3733
Shweta V. Bondre , Uma Yadav , Vipin D. Bondre , Poorva Agrawal
In India, agriculture is the primary source of income for half the people. Even in situations of fast population growth, agriculture supplies nourishment for all people. To provide food for the entire population, it is advised to detect plant diseases at an early stage. Plant leaf diseases are recognized using images of the affected leaves. Deep learning (DL) research seems to offer several opportunities for increased accuracy. Ant colony optimization with convolution-neural-network (ACO-CNN), a new deep learning technique for identifying and categorizing diseases, is presented in this article. Ant colony optimization (ACO) was used to examine the efficacy of disease diagnostics in plant leaves. The convolution neural network (CNN) classifier is used to remove texture, color, and leaf arrangement geometry from the input images. The ACO-CNN model outperformed the support vector machine (SVM) and CNN models in terms of precision, recall, and accuracy. CNN's rate is 81.6% as compared to SVM's 80% accuracy level. In the “ACO-CNN” approach, the F1-score, recall, and precision have higher rates as compared to other models, and the “F1-score” has the highest rate compared with other models since the ACO-CNN model has an accuracy rate of 91.00%.
Volume: 14
Issue: 5
Page: 3724-3733
Publish at: 2025-10-01

Multi-class stock market forecasting with deep learning models: an explainable artificial intelligence

10.11591/ijai.v14.i5.pp4342-4352
Chhaya Patel , Ashwin Raiyani
In this research, we investigated the influence of different deep learning techniques on time series stock market data, especially for all Nifty50 companies in the Indian stock market. Our proposed method of stock market prediction focused on multi-class classification with explainable artificial intelligence (XAI). Our proposed model incorporates convolutional neural network (CNN) for operational feature extraction and long short-term memory (LSTM) to capture time-based dependencies. Predicted value is classified with multiclass classes-very bullish, bullish, neutral, bearish, very bearish signals for all Nifty50 stocks. The model integrates essential technical indicators to find patterns from basic price data. XAI techniques are also used to find feature contributions to model prediction. It improves the clarity of the model’s administrative procedure by figuring out how technical indicators influence stock estimates. The outcomes highlight the model’s ability to generate actionable trading signals, reinforced by performance progress metrics, contributing to more well-informed and planned venture decisions. The proposed model reveals greater performance, reaching an average accuracy of 96%, beating LightGBM at 89%, random forest at 85%, and support vector machine at 60%.
Volume: 14
Issue: 5
Page: 4342-4352
Publish at: 2025-10-01

Translation-based image steganography system utilizing autoencoder and CycleGAN

10.11591/ijai.v14.i5.pp3958-3969
Thakwan Akram Jawad , Jamshid Bagherzadeh Mohasefi , Mohammed Salah Reda Abdelghany
Traditional image steganography involves embedding secret information into a cover image, a process that requires modification of the carrier and potentially leaves detectable marks. This paper proposes a novel method of coverless image steganography based on generative models. Initially, a CycleGAN model is constructed and trained to learn the features of different image domains. Subsequently, an Autoencoder model is trained using two sets of images to achieve a precise one-to-one mapping. Once the models are trained, the autoencoder is used on both the sender and receiver sides to convert the cover image (also known as the stego image) into the secret image and vice versa. The CycleGAN model is then utilized to enhance the visual quality of the images generated by the autoencoder. Experimental results demonstrate that this method not only effectively secures secret information transmission but also improves efficiency and increases the capacity for information hiding compared to similar methods.
Volume: 14
Issue: 5
Page: 3958-3969
Publish at: 2025-10-01

Transforming campus mobility: the DigiSticker system in digital parking solutions

10.11591/ijai.v14.i5.pp3667-3680
Rahman Md. Mojnur , Md. Tanjil Sarker , Wan-Noorshahida Mohd-Isa , Fahmid Al Farid , Md. Rakibul Hassan Sheam , Hezerul Abdul Karim , Gobbi Ramasamy , Aziah Ali
University digital parking systems have several benefits and solve many problems with traditional parking. Universities without a digital parking system face restricted parking, traffic congestion, inefficient space utilization, security issues, limited decision-making data, and diminished sustainability initiatives. This study paper discusses the benefits of digital parking systems and the drawbacks of traditional methods. By using technology to streamline university parking administration, the DigiSticker system offers an innovative solution. The DigiSticker system improves parking efficiency, convenience, and security for students, professors, and staff by delivering real-time parking information, assistance, and automated payments. This system has a user-friendly website and mobile app, fast registration, gate access management, security, user experience enhancement, and sustainability. Universities can improve student and staff parking experiences while improving parking management efficiency, security, and sustainability by meeting these requirements.
Volume: 14
Issue: 5
Page: 3667-3680
Publish at: 2025-10-01

Efficiency search: application of nature-inspired algorithms in artificial intelligence forecasting models

10.11591/ijai.v14.i5.pp3528-3541
José Rolando Neira Villar , Miguel Angel Cano Lengua
This study reviews how nature-inspired optimization algorithms (NIOAs) have been applied to artificial intelligence-based demand forecasting, using preferred reporting items for systematic reviews and meta-analyses (PRISMA) and clustering analysis to examine 36 selected articles. The findings reveal that NIOAs, particularly genetic algorithms and swarm intelligence methods, including their hybrids, have been frequently applied to long short-term memory (LSTM) and other backpropagation neural network models (BPNN). A key insight is the differentiated application of NIOAs depending on network depth: In shallow networks, they have been effectively used to optimize trainable parameters, whereas in deep networks, their role has focused primarily on hyperparameter optimization due to the prohibitive dimensionality of trainable weights. In all studies, NIOA-optimized models consistently outperform conventional baselines based on backpropagation. However, persistent challenges such as excessive execution times and slow convergence have led to the development of more efficient hybrid strategies and adaptive mechanisms for automated exploration-exploitation control. By mapping explored and unexplored pathways, summarizing key outcomes and techniques, and identifying promising methodologies, this review offers a practical foundation to guide future experiments and implementations involving NIOA-based optimization strategies in neural network models. As a conceptual contribution, it also proposes an innovative use of multispace optimization to address one of the most critical challenges identified: the optimization of trainable parameters in deep neural networks.
Volume: 14
Issue: 5
Page: 3528-3541
Publish at: 2025-10-01

Comparative analysis of convolutional neural network architectures for poultry meat classification

10.11591/ijai.v14.i5.pp3715-3723
Sekhra Salma , Mohammed Habib , Adil Tannouche , Youssef Ounejjar
The increasing demand for standardized food quality assurance, particularly in regions like Morocco, emphasizes the need for accurate classification of poultry meat. This study evaluates and compares ten convolutional neural network (CNN) architectures—VGG19, VGG16, ResNet50, GoogleNet, MobileNetV1, MobileNetV2, DenseNet, NasNet, EfficientNet, and AlexNet—for classifying commonly consumed poultry meat types in Moroccan markets, including chicken, turkey, fayoumi, and farmer’s chicken. A labeled image dataset was used to train and test each model, with performance assessed using metrics such as accuracy, precision, recall, training time, and computational complexity. Additionally, the study investigates how dataset size influences model performance, addressing challenges like limited data availability and scalability. The results highlight DenseNet as the top-performing architecture, achieving 98% classification accuracy while also demonstrating superior computational efficiency. These findings are valuable for improving food quality control, offering data-driven support for stakeholders in poultry production, distribution, and regulatory bodies. By identifying optimal deep learning models for poultry meat classification, the study contributes to enhancing food authentication and safety in Morocco and similar regions. It also encourages the integration of AI-driven systems in food inspection processes, providing scalable, accurate, and efficient solutions for ensuring standardized quality in the poultry supply chain.
Volume: 14
Issue: 5
Page: 3715-3723
Publish at: 2025-10-01

Optimizing battery life: a TinyML approach to lithium-ion battery health monitoring

10.11591/ijai.v14.i5.pp3858-3868
Kamaraj Lalitha Nisha , Vasanth Pradeep , Padmanabhan Puthiyaveedu Krishnankutty Nair , Sreelakshmi Pillai , Manikandan Arunachalam , Rakesh Thoppaen Suresh Babu
Electrical vehicles (EVs) are crucial nowadays due to their reduction in greenhouse gas emissions, decreasing dependence on remnant fuels, and improving air quality. For EVs, the battery is the heart that determines range, performance, and efficiency. Also, it directly impacts the cost and overall vehicle life span. Lithium-ion (Li-ion) batteries are pivotal in powering modern portable electronics and electric vehicles due to their high energy density and durability. Issues with current batteries include slow charging, short cycles, and low energy density. Most of the problems with current batteries are resolved by Li-ion batteries, which also helps explain why EV usage is increasing globally. However, to guarantee maximum performance and safety, estimating the remaining useful life and health state of these batteries remains a major difficulty. To improve battery lifetime of the battery and to overcome the problems of delayed charging, this study introduces a tiny machine learning (TinyML) method. An innovative machine learning approach is put forth that allows for effective learning on devices with limited resources, which enables real-time monitoring of the health status of the Li-ion batteries.
Volume: 14
Issue: 5
Page: 3858-3868
Publish at: 2025-10-01

Improving multilayer perceptron on rainfall data using modified genetics algorithm

10.11591/ijai.v14.i5.pp3994-4005
Marji Marji , Wayan Firdaus Mahmudi , Endang Wahyu Handamari , Edy Santoso , Maulana Muhamad Arifin
Rainfall prediction is essential for managing water resources, agriculture, and disaster response, particularly in regions affected by climate variability. This study introduces a modified genetic algorithm (MGA) to optimize hyperparameters of a multilayer perceptron (MLP) for rainfall forecasting. The MGA incorporates elitism to retain top-performing solutions and adaptive selection based on model accuracy. The proposed MGA–MLP model was tested on rainfall datasets from Australia and Indonesia (BMKG). Experimental results show that configurations with two hidden layers, rectified linear unit (ReLU) activation and limited-memory Broyden Fletcher Goldfarb Shannon (LBFGS) optimizer, a learning rate of 0.001 and 1000 epochs consistently delivered strong performance. The model achieved accuracies of 86.02% and 79.05%, respectively. These findings indicate that MGA significantly improves MLP performance and provides a reliable, generalizable method for rainfall prediction across diverse climatic conditions.
Volume: 14
Issue: 5
Page: 3994-4005
Publish at: 2025-10-01

Ensemble reverse knowledge distillation: training robust model using weak models

10.11591/ijai.v14.i5.pp4162-4170
Christopher Gavra Reswara , Tjeng Wawan Cenggoro
To ensure that artificial intelligence (AI) can be aligned with humans, AI models need to be developed and supervised by humans. Unfortunately, it is possible for an AI to exceed human capabilities, which is commonly referred to as superalignment models. Thus, it raised the question of whether humans can still supervise a superalignment model, which is encapsulated in a concept called weak-to-strong generalization. To address this issue, we introduce ensemble reverse knowledge distillation (ERKD), which leverages two weaker models to supervise a more robust model. This technique is a potential solution for humans to manage a super-alignment of models. ERKD enables a more robust model to achieve optimal performance with the assistance of two weaker models. We tried to train a more robust EfficientNet model with weaker convolutional neural network (CNN) models in a supervised fashion. With this method, the EfficientNet model performed better than the model trained with the standard transfer learning (STL) method. It also performed better than a model that was supervised by a single weaker model. Finally, ERKD-trained EfficientNet models can perform better than EfficientNet models that are one or even two levels stronger.
Volume: 14
Issue: 5
Page: 4162-4170
Publish at: 2025-10-01

BonoNet: a deep convolutional neural network for recognizing bangla compound characters

10.11591/ijai.v14.i5.pp4171-4180
Kazi Rifat Ahmed , Nusrat Jahan , Adiba Masud , Nusrat Tasnim , Sazia Sharmin , Nusrat Jahan Mim , Imran Mahmud
The bangla alphabet includes vowels, consonants, and compound symbols. The compound nature of bangla is a product of combining two or more root bangla characters into one graph. They are difficult to differentiate because they have a sophisticated geometric shape and an immense variety of scripts used by different places and individuals. This is one of the greatest challenges in creating effective optical character recognition (OCR) systems for bangla. In this paper, a deep convolutional neural network (DCNN)-based system is presented to identify bangla compound characters with high precision. The model was trained using the AIBangla dataset. It has about 171 classes of bangla compound characters. A DCNN system, BonoNet, was designed to classify compound characters. BonoNet outperformed all other state-of-the-art architecture on the test set and improved over current state-of-the-art architecture methods. BonoNet will greatly improve the automation and analysis of the bangla language by accurately identifying these compound complex characters.
Volume: 14
Issue: 5
Page: 4171-4180
Publish at: 2025-10-01
1 2 3
...
2097 2098
Show 217 of 2098

Discover Our Library

Embark on a journey through our expansive collection of articles and let curiosity lead your path to innovation.

Explore Now
Library 3D Ilustration