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

31,377 Article Results

Hyperparameter optimization of deep residual recurrent fusion models for facial emotion recognition

10.11591/ijai.v15.i3.pp2581-2594
Muhammad Munsarif , Ku Ruhana Ku-Mahamud
Deep learning facial emotion recognition (FER) is widely applied in healthcare, education, and human–computer interaction. However, many deep learning models suffer from suboptimal hyperparameter configurations that reduce accuracy and stability. This study proposes three deep residual recurrent fusion models that integrates residual blocks with recurrent neural networks (bidirectional long short-term memory (BiLSTM), long short-term memory (LSTM), and gated recurrent unit (GRU)) to capture both spatial and temporal features. A systematic hyperparameter optimization strategy was applied, tuning kernel size, filter size, recurrent units, batch size, learning rate, dropout, and weight decay to balance generalization and computational efficiency. The models were evaluated on four benchmark datasets: FER2013, FERPlus, RAF-DB, and CK+. The results show that optimized configurations achieved outstanding accuracy, reaching 99.85% on FER2013, 99.99% on FERPlus, and 100% on RAF-DB and CK+. These findings demonstrate that careful hyperparameter tuning significantly enhances feature extraction, mitigates vanishing gradient and overfitting issues, and improves generalization across diverse datasets. The proposed framework highlights the importance of optimization in advancing robust FER systems for real-world applications.
Volume: 15
Issue: 3
Page: 2581-2594
Publish at: 2026-06-01

AI-driven log reduction and storage optimization for security operations

10.11591/ijece.v16i3.pp1417-1424
Nutthakorn Chalaemwongwan
In this study, we present an AI-driven framework that integrates semantic log reduction with compliance-aware storage optimization, specifically designed for security operations center (SOC) and managed security service provider (MSSP) environments. Traditional approaches such as uniform compression, keyword filtering, and static tiering often either miss critical anomalies or preserve redundant noise, leading to excessive storage use, slower search performance, and analyst fatigue. The proposed framework addresses these challenges by combining three components: semantic reduction of repetitive entries, anomaly-focused retention supported by self-supervised models, and adaptive tiering aligned with regulatory requirements. Evaluations on HDFS, BGL, CICIDS2017, and Suricata datasets achieved 70%–80% log reduction, 55%–65% storage savings, recall rates above 95%, and a one-third reduction in query latency. These results demonstrate that pre-index reduction, together with anomaly- and compliance-aware retention, offers a scalable and regulator-ready solution for operational security environments.
Volume: 16
Issue: 3
Page: 1417-1424
Publish at: 2026-06-01

A scalable low-cost internet of things-based electronic nose for identifying chemical ripening in fruits

10.11591/ijai.v15.i3.pp2113-2124
Amit R. Welekar , Gaurav Vishnu Londhe , Amit Pimpalkar , Abhrendu Bhattacharya , Nilesh Shelke , R. Jeberson Retna Raj
Indiscriminate use of chemical agents like calcium carbide and ethephon for the ripening of fruits poses grave health hazards, emitting carcinogenic and neurotoxic compounds. Here we present a new, scalable, inexpensive, internet of things (IoT)-enabled electronic nose (e-nose) AI-Bot system for the detection of chemically ripened fruits. This would involve the development of a system that uses an MQ-3 gas sensor to quantify the ethanol content, as well as an MQ-135 gas sensor with an ESP32 microcontroller to quantify even further the amount of volatile organic compounds (VOCs) suggestive of artificial ripening. Flutter-based mobile application allows real-time monitoring, ripening classification using machine learning (ML) algorithms, and logging the historical data. A small sample was taken for inter-document feature literature mining, modelling sensor behaviors according to voltage dividers and gas concentration resistance laws for robust calibration and classification performance. Validation studies were performed on mango, banana, and papaya fruit in the laboratory environment. Total 75 samples (25 each of banana, mango, papaya across 3 trials) of fruit were tested. The implemented system achieved 95% for banana, 92% for mango, and 90% classification accuracy for papaya when cross validated.
Volume: 15
Issue: 3
Page: 2113-2124
Publish at: 2026-06-01

Multi-agent autonomous GeoAI framework for scalable and self-improving geospatial intelligence

10.11591/ijai.v15.i3.pp2201-2215
Kim-Son Nguyen , The-Vinh Nguyen , Van-Viet Nguyen , Thi-Minh-Hue Luong , Huu-Khanh Nguyen , Duc-Binh Nguyen
Large language models (LLMs) have recently expanded the scope of automation across many application domains. In geographic information systems (GIS), however, many tasks still require specialized expertise and remain difficult for non-expert users. Recent studies have explored LLM-based geospatial analysis under a single-agent paradigm, but these early systems remain limited by weak coordination, limited error recovery, and dependence on proprietary artifacts. This study proposes multi-agent autonomous geospatial artificial intelligence (MA-GeoAI), a multi-agent architecture in which the planner, coder, validator, debugger, and knowledge agents collaborate through the LangGraph framework. The framework was evaluated on three case studies: population exposure assessment, mobility pattern analysis, and county-level mortality modeling. Unlike general-purpose multi-agent LLM frameworks, MA-GeoAI embeds spatial semantics, coordinate reference system (CRS) consistency checks, geometry validation, and operation-aware coordination directly into the control loop. Across repeated runs, all evaluated systems completed the controlled artifact contract; therefore, the analysis focuses on auditability, runtime, fallback behavior, and reproducibility rather than binary task-completion superiority.
Volume: 15
Issue: 3
Page: 2201-2215
Publish at: 2026-06-01

A transfer hybrid deep learning approach for advanced intrusion detection in IoT-based smart home security

10.11591/ijai.v15.i3.pp2750-2760
Mouad Choukhairi , Ouail Choukhairi , Youssef Fakhri , Ali Choukri
As smart home environments increasingly rely on interconnected internet of things (IoT) devices, they face growing cyber threats originating both externally from malicious actors and internally from compromised or malfunctioning IoT devices. These threats, including unauthorized access, distributed denial of service (DDoS) attacks, and data exfiltration, pose significant risks to the security and privacy of smart home inhabitants. This paper introduces an advanced intrusion detection system (IDS) specifically designed to enhance the security of IoT-based smart home networks. Leveraging a hybrid deep learning approach combining convolutional neural networks (CNN) and long short-term memory (LSTM) models, complemented by transfer learning (TL) and hyper-parameter optimization techniques, our proposed IDS efficiently identifies both external and intra-network threats. Using the IoTID20 dataset, which simulates realistic attack scenarios, the IDS was trained and evaluated to detect abnormal behavior effectively within smart home networks. CNN layers extract spatial features from network traffic, while LSTM layers capture temporal dependencies, enabling robust detection against a range of cyber-threats. Evaluation results demonstrate the IDS’s high detection accuracy and exceptional F1-scores, validating its effectiveness in safeguarding IoT-based smart homes from evolving threats.
Volume: 15
Issue: 3
Page: 2750-2760
Publish at: 2026-06-01

Class imbalance resolution in IoT networks using advanced elk herd optimization with SMOTE and iteratively fine-tuned deep BiLSTM

10.11591/ijai.v15.i3.pp2920-2934
Srikanth Mudiyanur Sriramappa , Ananda Babu Jayachandra , Vasantha Kumar Mahadevachar , Ashwini Kailas , T. G. Keerthan Kumar
The "internet of things (IoT)" mentions to a system in which multiple network protocols are used to link together disparate devices that are always sharing information with one another. The hope is that this study will add to the existing body of knowledge by suggesting ways to make intrusion detection systems (IDS) more effective. Training datasets for the proposed model were collected from telemetry of network (ToN)-IoT network traffic. After cleaning and normalizing the datasets, the synthetic minority over sampling technique (SMOTE) is used to balance the datasets that are imbalanced. Optimal sampling rates are critical for resolving class imbalance, as SMOTE's efficiency is dependent on it for instances involving minority classes. Improving classification accuracy through finding appropriate sample rates for input datasets is the goal of this paper's introduction of advanced elk herd optimization (AEHO) with SMOTE. Finally, a deep bidirectional long short-term memory (deep BiLSTM) model based on deep learning is used to classify attacks. The fine-tuning technique is used during testing to update the high limits and when combined with the data balancing mechanism and AEHO-based-SMOTE, the results greatly enhance the classification techniques. Deep BiLSTM performs better than 90% in every category: classification, recall, precision, and F1-score.
Volume: 15
Issue: 3
Page: 2920-2934
Publish at: 2026-06-01

RGB-D salient object detection with local feature and semantic segmentation

10.11591/ijai.v15.i3.pp2774-2785
Zhang Wang , Kim On Chin , Rayner Alfred , Junyi Chai , Rundong Zhang , Soo See Chai
Red, green, blue–depth (RGB-D) salient object detection (SOD) focuses on identifying visually prominent objects by simulating human visual perception. While existing RGB-D SOD methods have demonstrated results, there remain challenges in effectively leveraging extrinsic cues and enhancing feature representation. To address these limitations, novel RGB-D SOD model with local feature extraction and semantic segmentation (LFSS) is introduced, which is built on an encoder-decoder architecture. The encoder preprocesses the input images by merging RGB and depth data through a channel and spatial attention (CSA) module. A local feature extraction module further refines this fusion. The decoder consists of three key modules: i) the multi-feature extraction (MFE) module enhances base features through diverse convolutional operations; ii) the semantic segmentation enhancement (SSE) module optimizes features via spatial pyramid pooling and atrous convolution; and iii) the local/global agreement and edge detection (LGE) module that enables multi-level feature interaction and edge detection. These modules work sequentially to enhance and extract salient objects. LFSS is evaluated on six standard RGB-D SOD datasets (NJU2K, NLPR, STERE, LFSD, SSD, SIP) by four metrics, outperforming the comparison models with up to 1.2% F-measure improvement. LFSS is found to be a versatile model, offering valuable applications in engineering.
Volume: 15
Issue: 3
Page: 2774-2785
Publish at: 2026-06-01

A novel deep learning-based pollinator species classification using the multimodal species-image regularization framework

10.11591/ijai.v15.i3.pp2685-2697
Pooja Hadimane , Ashoka Kukkuvada
Pollinators, such as bees, butterflies, and other insects, are essential to maintaining biodiversity and ensuring agricultural productivity, with over 80% of flowering plants and 75% of global food crops relying on them for successful reproduction. However, pollinator populations are facing significant declines due to environmental changes, habitat destruction, and climate change, posing substantial risks to ecosystems and global food security. This paper introduces the multimodal species-image regularization (MSIR) framework for automating the classification of pollinator species using both binary classification (pollinator vs. non-pollinator) and multiclass classification (bee genera). The framework integrates multimodal data, including visual images of pollinators and species details such as genus, family, and environmental factors, to improve accuracy and scalability. The system leverages the multimodal contrastive learning framework (MCLF) to align both image and species-detail features into a unified embedding space, enabling mor effective classification. Additionally, the framework applies image-species prototype regularization (ISPR) and species-detail prototype regularization (SDPR) to further enhance the classification accuracy by regularizing the tunable weights based on prototype alignment. The proposed deep learning (DL) model is evaluated against traditional machine learning (ML) methods, such as random forest (RF), and demonstrates superior performance on key metrics, including accuracy, precision, recall, and F1-score.
Volume: 15
Issue: 3
Page: 2685-2697
Publish at: 2026-06-01

Computed tomography imaging radiomics: a novel approach to early-stage non-small cell lung cancer prediction

10.11591/ijai.v15.i3.pp2471-2483
Raviteja Balekai , Mallikarjun S. Holi
Radiomics shows promise as non-invasive method for enhancing clinical staging of non-small cell lung cancer (NSCLC) by using quantitative information from computed tomography (CT) scans. This study presents radiomics-based machine learning (ML) approach for staging NSCLC patients into clinical stages I, II, and III based on shape, intensity, and texture features. CT images of 369 NSCLC patients are collected from the cancer imaging archive (TCIA), and extracted 107 radiomic features following image biomarker standardization initiative (IBSI) protocol. The analysis of the sources of variability due to different imaging protocols, using principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), showed that these effects were resolved through ComBat harmonization. Recursive feature elimination (RFE) and least absolute shrinkage and selection operator (LASSO) are used for feature selection. Five ML algorithms: logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost) were used, with an 80:20 train-test split and 10-fold cross-validation. The classifier is assessed using accuracy, sensitivity, specificity, F1-score, and area under the receiver operating characteristic (AUROC) curve. The RFE and RF classifier combination performed the best with AUROC of 0.9307 and accuracy of 0.8114. This study illustrates the use of radiomics models in non-invasive classification of NSCLC stages and it is role in clinical decision making.
Volume: 15
Issue: 3
Page: 2471-2483
Publish at: 2026-06-01

Optimized ResNet-50 framework for mammogram-based breast cancer classification: a comparative evaluation with EfficientNet-B0

10.11591/ijece.v16i3.pp1199-1212
Muhammad Subali , Lulu Mawaddah Wisudawati , Teresa Teresa
Breast cancer remains one of the most prevalent malignancies worldwide, underscoring the need for accurate and reliable mammographic interpretation. Computer-aided diagnosis (CAD) based on deep learning has emerged as a promising approach to improve both screening performance and diagnostic consistency, yet fairness-driven comparisons between popular convolutional backbones on public mammogram benchmarks remain limited. This study provides a statistically validated, fairness-driven comparison of two widely used convolutional neural network architectures, ResNet-50 and EfficientNet-B0, for mammogram-based breast cancer classification under a rigorously controlled, clinically motivated protocol. The proposed “optimized ResNet-50” framework is defined by patient-level stratified undersampling, paired 5-fold cross-validation with identical partitions, harmonized augmentation and training configurations, and dual statistical testing (paired t-tests and Wilcoxon signed-rank tests), emphasizing methodological rigor rather than architectural novelty. Across MIAS and CBIS-DDSM benchmarks, the models demonstrated complementary strengths, with EfficientNet-B0 excelling in screening-oriented tasks (normal vs. abnormal) and ResNet-50 offering more robust performance for diagnostic-oriented tasks (benign vs. malignant). These findings highlight the value of fairness-driven evaluation protocols in CAD research and support the feasibility of integrating lightweight convolutional neural networks (CNNs) into tiered clinical workflows, where different backbones are strategically deployed for initial screening and confirmatory assessment.
Volume: 16
Issue: 3
Page: 1199-1212
Publish at: 2026-06-01

Exploring artificial intelligence in vocational learning: teachers’ perspective from Indonesia

10.11591/ijai.v15.i3.pp2009-2023
Yuliansah Yuliansah , Mar’atus Sholikah , Sutirman Sutirman
The rapid development of artificial intelligence (AI) in education has expanded its potential applications. However, empirical evidence from vocational education in developing countries remains limited, particularly regarding differences between certified and uncertified vocational high school (VHS) teachers’ perspectives. This study investigates VHS teachers’ perceptions of AI use in the learning process by explicitly comparing certified and uncertified teachers in Indonesia. Using a quantitative approach combined with data-mining techniques applied to open-ended survey responses, data were collected from 65 VHS teachers in the Special Region of Yogyakarta (DIY). Group differences were examined using a non parametric Mann-Whitney U test. The findings indicate that both certified and uncertified teachers demonstrate consistently positive perceptions of AI in instructional planning, implementation, and assessment, with no statistically significant differences between the two groups. Importantly, this result suggests that openness toward AI integration is not determined by certification status but reflects broader pedagogical orientations among vocational teachers. Teachers perceive AI primarily as a pedagogical partner rather than a substitute for professional educators. The study underscores the need of structured AI-focused professional development and policy support through adequate infrastructure and targeted training to enhance the effectiveness of AI adoption in improving the quality of vocational education in Indonesia.
Volume: 15
Issue: 3
Page: 2009-2023
Publish at: 2026-06-01

A hybrid deep learning approach for BoT-IoT intrusion detection

10.11591/ijai.v15.i3.pp2192-2200
Khalid Altarawneh , Ghayth AlMahadin , Ibrahim Altarawni
Internet of things (IoT) devices enhance quality of life and industrial operations but pose significant security risks, necessitating intelligent intrusion detection systems (IDS) to combat evolving cyber threats. This paper proposes a novel IDS framework integrating bio-inspired heuristic feature selection, a generative adversarial network (GAN)-based data augmentation, and an ensemble classifier combining ResNet, AlexNet, and MobileNet. The methodology, tested on the botnet (BoT)-IoT dataset, follows four stages: preprocessing, feature augmentation, feature selection, and ensemble classification. Evaluated on benchmarks including CIC-IDS-2018, NSL-KDD, and UNSW-NB15, the model achieved accuracies of 98.2%, 99.1%, 97.6%, and 98.4%, respectively, with consistently high precision, recall, and F1-scores, demonstrating robust detection of diverse cyberattacks. Beyond accuracy, the framework optimizes processing time for large-scale IoT data, addressing scalability challenges in real-time threat mitigation. By synergizing feature optimization, synthetic data generation, and deep learning architectures, the solution enhances detection rates while minimizing computational overhead. Comparative analysis highlights its superior performance over existing methods, positioning it as a vital tool for securing IoT ecosystems against unauthorized access and malicious activities. The results underscore its potential to fortify IoT network security, balancing efficiency, adaptability, and computational feasibility for practical deployment in resource-constrained environments.
Volume: 15
Issue: 3
Page: 2192-2200
Publish at: 2026-06-01

Comparing global and variety-specific ensemble models for avocado maturity prediction with near-infrared

10.11591/ijai.v15.i3.pp2300-2313
Christian Ovalle , Jhon Garcia Jimenez , Jose Briones Zuñiga
Ensuring accurate, non-destructive maturity classification of avocados is critical to supply chain optimization in agro-industrial systems. This study presents a predictive framework that integrates near-infrared (NIR) spectroscopy with ensemble stacking machine learning (ML) to enhance the precision of avocado ripeness assessment. The proposed methodology compares global versus variety-specific models for 'Hass' and 'Fuerte' avocado types, leveraging spectral data (900–1,700 nm) and multiple base classifiers, including random forest (RF), gradient boosting (GB), support vector machines (SVMs), decision trees (DT), k-nearest neighbors (KNN), and categorical boosting (CatBoost), combined via linear regression as a meta-learner. Experimental results revealed that the stacking models outperformed individual learners, with variety-specific GB model achieving the highest performance (Matthews correlation coefficient (MCC) =0.679, area under the curve (AUC) =0.931). These findings highlight the critical importance of varietal specificity in model calibration and demonstrate how ensemble strategies can improve robustness, scalability, and interpretability in intelligent agricultural systems. The proposed model provides a computationally efficient solution for real-time quality control and supports the deployment of AI-powered systems within agricultural supply chains in developing regions.
Volume: 15
Issue: 3
Page: 2300-2313
Publish at: 2026-06-01

Enhancing fake news detection: a hybrid BERT-XGBoost model for improved performance and interpretability

10.11591/ijai.v15.i3.pp2385-2397
Nishant Vasantkumar Hegde , Suneesh Bare , Namruth Reddy , Rajat Gondkar Aravinda , Minal Moharir , Aamir Ibrahim
The widespread spread of fake news poses a serious threat to the integrity of information. The dominant approach to detection involves end-to-end fine-tuning of large transformer models like bidirectional encoder representations from transformers (BERT), which, despite achieving high accuracy, often function as opaque “black boxes” with limited interpretability. This paper proposes and validates a hybrid, decoupled architecture that proves to be a more practical and powerful alternative. We first fine-tune a DistilBERT model on the full WELFake dataset of 71,537 articles after cleaning to create domain-specific embeddings. These high-dimensional vectors are then used as input features to train a robust extreme gradient boosting (XGBoost) classifier. The results demonstrate that the hybrid model achieves a state-of-the-art accuracy of 99.76%, slightly surpassing the already high performance of a standard end-to-end fine-tuned model. Crucially, this approach provides this top-tier performance while offering significant advantages in model interpretability through feature importance analysis. This work establishes that a decoupled architecture is not just a viable alternative but a superior practical strategy for combating misinformation, successfully balancing state-of-the-art accuracy with essential model transparency.
Volume: 15
Issue: 3
Page: 2385-2397
Publish at: 2026-06-01

Double direction optimization: a new metaheuristic that performs exploitation and exploration simultaneously

10.11591/ijai.v15.i3.pp2874-2884
Purba Daru Kusuma , Helmy Widyantara
This research constructs a novel method called double direction optimization (DDO). DDO is constructed based on swarm intelligence (SI) approach and it does not use any metaphor. As its name suggests, it employs a novel algorithm by performing exploitation and exploration simultaneously which is transformed into two sequential searches. In the 1st search, the motion toward the highest quality agent is combined with the motion toward a randomly taken higher quality agent. In the 2nd search, the motion toward the finest entity is combined with the motion relative to a randomly taken agent. In this work, the efficacy of the DDO is assessed using three use cases: 23 functions, four engineering problems, and an economic emission dispatch (EED) problem. In this assessment, there are five metaheuristics that become the benchmark: crayfish optimization algorithm (COA), hiking optimization (HO), osprey optimization algorithm (OOA), carpet weaver optimization (CWO), and dollmaker optimization algorithm (DOA). The result indicates the supremacy of DDO in high dimension functions and competitiveness of DDO in fixed dimension multimodal functions, four engineering problems, and the EED problem.
Volume: 15
Issue: 3
Page: 2874-2884
Publish at: 2026-06-01
1 2 3
...
2091 2092
Show 66 of 2092

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