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

An explainable artificial intelligence benchmarking with deep reinforcement learning for telecom battery backup prediction

10.11591/ijai.v15.i5.pp4482-4496
Promphak Boonraksa , Kedsara Palachai , Veerapatra Wangsilabatra , Terapong Boonraksa
Accurate prediction of battery backup time is essential for ensuring service reliability and efficient energy management in modern telecommunication networks, particularly during power outages. This study proposes a comprehensive benchmarking framework to evaluate the performance of eleven advanced time-series forecasting models, including deep learning architectures (multi-layer perceptron (MLP), convolutional neural network 1-dimension (CNN1D), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), gated recurrent unit (GRU), and temporal convolutional network (TCN)), Transformer-based models (Transformer, Informer-Lite, and patch time-series Transformer-lite (PatchTST-Lite)), and linear baseline approaches (decomposition linear (DLinear) and normalized linear (NLinear)). The models are trained and validated using real-world telecom power system data under varying load and environmental conditions. To improve model transparency and reliability, an explainable artificial intelligence (XAI) framework is incorporated to interpret feature contributions and identify key factors influencing battery performance. Experimental results indicate that Transformer-based models outperform conventional deep learning and linear models in terms of prediction accuracy and robustness across different operating scenarios. Furthermore, XAI analysis highlights that load current, ambient temperature, and battery health indicators are the most influential variables affecting backup time prediction. The proposed framework provides both high accuracy forecasting and interpretability, enabling more effective decision-making and enhancing energy resilience in telecommunication infrastructure.
Volume: 15
Issue: 5
Page: 4482-4496
Publish at: 2026-10-01

Autoencoder-based anomaly detection in water cyber-physical systems

10.11591/ijai.v15.i5.pp4635-4644
Sara Tahiri , Mouad Choukhairi , Youssef Fakhiri , Mohamed Amnai
Cyber-physical systems (CPS) are at the core of enabling services in the critical infrastructures, such as a smart water network, hence the need for effective anomaly detection to ensure operational safety and security. This paper proposes deep water anomaly detection (DeepWAD), a big data-based multiple purpose anomaly detection tool that leverages deep learning, supports big data, and internet of things (IoT) data collection and analysis. In a unified processing pipeline, it estimates and contrasts three kinds of autoencoders, viz adversarial autoencoder (AAE), variational autoencoder (VAE), and classical autoencoder (CAE). These models are then applied to a practical water treatment system, namely the secure water treatment (SWaT) dataset. Using reconstruction errors, DeepWAD analyzes time-series data generated by IoT sensors in order to discover anomalies including, for example, equipment malfunctions, sensor errors, and cyber threats. The framework also demonstrates the performance of various architectures in terms of data complexity, training consistency, and reconstruction accuracy. Due to its generality, DeepWAD can also be applied in other CPS domains such as the energy grid or industrial control systems. It serves as a decision-support tool to select the most appropriate anomaly detection model depending on system constraints and detection requirements.
Volume: 15
Issue: 5
Page: 4635-4644
Publish at: 2026-10-01

Quantitative chest computed tomography-based hybrid deep learning for chronic lung disease detection and staging

10.11591/ijai.v15.i5.pp4568-4591
Garima Jain , Amit Kumar Goel , Rahul Kumar
This research proposes a hybrid deep learning model to accurately detect chronic lung diseases in chest computed tomography (CT) scans. Current deep learning methods, such as convolutional neural networks (CNNs) and transformer models, often suffer from high computational requirements and low interpretability. This research combines a pre-trained residual network 50 layers (ResNet50) for deep feature extraction with hand-crafted texture features. The features are combined via hybrid concatenation, and the model is classified using an optimized extreme gradient boosting (XGBoost) ensemble with hyperparameter tuning to improve performance and reduce overfitting. This research compared the proposed model with radiomics-only, ResNet50 CNN, vision transformer (ViT-B16), and Swin transformer methods. This research used a public dataset containing adenocarcinoma, squamous cell carcinoma, large cell carcinoma, and normal classes. The classification accuracy of the hybrid framework was 92.38%, with better precision, sensitivity, specificity, F1-score, and receiver operating characteristic-area under the curve (ROC-AUC) than the baselines. A dual-mode explainable artificial intelligence (XAI) framework integrating gradient-weighted class activation mapping (Grad-CAM) and Shapley additive explanation (SHAP) was implemented to enable spatial lesion localization and feature attribution for clinical transparency. Grad-CAM was able to localize the lesions successfully, and SHAP revealed the influence of entropy-based texture features and semantic embeddings on the predictions. Overall, the framework develops advanced XAI systems for chest CT analysis.
Volume: 15
Issue: 5
Page: 4568-4591
Publish at: 2026-10-01

Stacking ensemble for multilingual aspect-based sentiment analysis of tourism reviews

10.11591/ijai.v15.i5.pp4381-4391
Basworo Ardi Pramono , Rahmat Gernowo , Aghus Sofwan
Multilingual tourism reviews often express opinions about multiple destination attributes within a single text, making aspect-level sentiment classification challenging. This study compares classical machine learning and ensemble methods for multilabel aspect-based sentiment analysis of amenities, attractions, and accessibility. The dataset comprises 13,085 Indonesian- and English-language Google Reviews labeled through a rule-based procedure. Reviews were represented using term frequency-inverse document frequency features and classified using logistic regression, multinomial naive Bayes, random forest, soft voting, adaptive boosting (AdaBoost), and stacking under a one-vs-rest framework. The stacking classifier combined logistic regression and random forest as base learners with logistic regression as the meta-learner. It achieved the best performance among the evaluated models, with a micro-F1 of 0.8626, macro-F1 of 0.7285, Hamming loss of 0.0633, and subset accuracy of 0.6825. These findings indicate that stacking effectively integrates complementary decision patterns and provides a reproducible framework for multilingual multilabel analysis of tourism reviews.
Volume: 15
Issue: 5
Page: 4381-4391
Publish at: 2026-10-01

MatchIA: a semantic artificial intelligence solution for interoperability in schema matching

10.11591/ijai.v15.i5.pp4213-4223
Mohamed Raoui , Moulay Hafid El Yazidi , Ahmed Zellou
Schema matching plays a crucial role in data integration by aligning attributes from heterogeneous sources. Traditional rule-based and statistical techniques often lack the flexibility to adapt to diverse schema structures, requiring extensive manual effort. To address these challenges, this paper introduces MatchIA, a machine learning solution that combines word embeddings and random forest to enhance the accuracy of schema matching. The approach accounts for both structural dependencies and semantic relationships between attributes, enabling more automated, and reliable correspondence identification. Unlike traditional methods, MatchIA leverages pretrained embeddings to capture contextual similarities and integrates learned patterns to generalize across various domains. Experimental evaluations on standard datasets demonstrate that MatchIA achieves a precision of 0.94, a recall of 0.92, and an overall matching accuracy of 0.93, significantly outperforming state-of-the-art methods. These results highlight the potential of MatchIA for real-world applications requiring scalable and automated data interoperability.
Volume: 15
Issue: 5
Page: 4213-4223
Publish at: 2026-10-01

Exploring Arabic non-sibilant fricative consonants: a comprehensive classification

10.11591/ijai.v15.i5.pp4878-4892
Youssef Elfahm , Nesrine Abajaddi , Badia Mounir , Laila Elmaazouzi , Ilham Mounir , Abdelmajid Farchi
Voice interaction serves as an efficient tool for improving communication between humans and machines, enabling quicker exchanges without the need for physical contact with devices. This study introduces a methodology aimed at identifying Arabic non-sibilant fricative consonants and recognizing their significance within this context. Our approach involves analyzing the normalized energy distribution within consonant-vowel phonetic syllables, which serves as the acoustic input signal for our speech recognition system. Notably, the relatively limited attention given to non-sibilant consonants in existing literature underscores the importance of this research endeavor. The evaluation of our classification method applied to our dataset highlights the pivotal role of vocal signal energy in characterizing Arabic non-sibilant fricative consonants. Achieving a classification accuracy of 90% for distinguishing voiced and voiceless non-sibilant consonants, with a rate exceeding 85% for other categories, underscores the effectiveness of our approach. Noteworthy is the outperformance of our classification technique compared to current algorithms based on support vector machines (SVM) and artificial neural networks (ANNs), indicating its potential to enhance Arabic automatic speech recognition (ASR) systems. Overall, this research contributes to advancing speech recognition technology, particularly in the realm of Arabic linguistics, offering promising implications for future developments in this field.
Volume: 15
Issue: 5
Page: 4878-4892
Publish at: 2026-10-01

Confidence-driven feature selection and ensemble classification using ResNet, AlexNet, and DenseNet for scene recognition

10.11591/ijai.v15.i5.pp4523-4539
Sumathi Krishnamurthy , Shetty Pramod Kumar , Halasur Rudrappa Mahadevaswamy , Ujwala Bagepalli Satishbabu
Indoor scene recognition is a significant problem because of its structural complexity, illumination variations, and visual similarities between the various scene classes. Recent approaches use convolutional neural networks (CNNs) and transfer learning to extract features for classification. In these methods, all the features were considered for classification, which leads to an increase in computational complexity and training time. This leads to more time consumption for processing. To address these challenges, this study proposed a hybrid indoor scene recognition framework that integrates illumination-aware pre-processing, multi-model deep feature extraction, and confidence-based feature fusion. The pre-processing technique uses contrast-limited adaptive histogram equalization (CLAHE) and Retinex enhancement. Deep features are extracted using Alex Krizhevsky network (AlexNet), densely connected convolutional network (DenseNet), and residual network 101 layers (ResNet101) to capture spatial and semantic information. The features extracted are fed to confidence-driven world cup optimization (CD-WCO) to extract the significant features. The final representation is classified using a deep liquid state machine (DLSM) classifier to enhance class discrimination. Our experiments, conducted on the NYU indoor scene dataset with 20 scene categories and the MIT Indoor-15 class subset results indicate the effectiveness of the proposed framework, achieving an overall classification accuracy of 96.5% and 95.2%, respectively.
Volume: 15
Issue: 5
Page: 4523-4539
Publish at: 2026-10-01

Performance analysis of convolutional neural network-based image classifiers for agricultural farmland mapping

10.11591/ijai.v15.i5.pp4403-4411
Anjela C. Tolentino , Anazel P. Gamilla , Amir Ledesma
Precise farmland mapping is essential for crop monitoring, land-use inventory, and subsidy audits. Particularly, the office of the municipality’s assessor or the registry of deeds needs adequate image classifiers to categorize farmland on a large scale across different seasons and regions. The study evaluated three convolutional neural network (CNN) models, such as efficient network baseline 3 (EfficientNetB3), residual network 50 layers version 2 (ResNet50V2), and inception version 3 (InceptionV3), to identify the model that best balances accuracy and efficiency for classifying high-resolution farmland imagery. EfficientNetB3 proved to be the most suitable model, demonstrating higher accuracy, robustness, and a stronger ability to identify rice farmland patterns. While ResNet50V2 and InceptionV3 are widely used for image analysis, their performance declined when processing farmland images that exhibit substantial differences in visual characteristics. EfficientNetB3 delivered more consistent results for farmland image classification in large-scale agricultural applications. These findings highlight its potential for supporting data-driven agricultural management and policy implementation.
Volume: 15
Issue: 5
Page: 4403-4411
Publish at: 2026-10-01

A unified artificial intelligence-enabled intelligent home assistant robot for smart automation and safety

10.12928/telkomnika.v24i5.27862
Md. Taslim; University of Frontier Technology Bangladesh (UFTB) Arif , Abu Salyh; University of Frontier Technology Bangladesh (UFTB) Muhammad Mussa , Jul Jalal; Daffodil International University (DIU) Al-Mamur Sayor , Md.; University of Frontier Technology Bangladesh (UFTB) Ashiqussalehin , Md. Mehedi; University of Frontier Technology Bangladesh (UFTB) Hasan , Mst Deloara; University of Frontier Technology Bangladesh (UFTB) Khushi , Md. Moshiur; University of Frontier Technology Bangladesh (UFTB) Rahman
Intelligent home assistant systems are increasingly required to enhance safety, efficiency, and user convenience in smart environments. This paper presents LUNA, a unified artificial intelligence (AI)-enabled home assistant robot de signed for smart automation and safety in indoor settings. The system integrates autonomous navigation, You Only LookOnceversion8(YOLOv8)-basedobject detection, gesture-based interaction, real-time gas and fire hazard detection, mo bile application control, and automated floor cleaning within a single low-cost embedded platform. Unlike existing systems that focus on isolated functional ities, the proposed approach combines perception, mobility, safety monitoring, and user interaction in a coordinated framework suitable for real-world appli cations, particularly for elderly and disabled users. The system is implemented using Arduino, Raspberry Pi, and ESP32 to enable efficient sensing, processing, communication, and actuation. Experimental results demonstrate robust perfor mance, achieving 93.5% navigation accuracy, 91.2% object detection accuracy, 2.5 s emergency response time, 1.2 s control latency, and 88.7% cleaning effi ciency. These results indicate that LUNA is an effective and practical solution for next-generation smart home assistance.
Volume: 24
Issue: 5
Page: 1680-1689
Publish at: 2026-10-01

Machine learning-based joint optimization for LoRaWAN reliability and energy efficiency

10.12928/telkomnika.v24i5.27916
Nana; University of Indonesia Rusdiana , Novi; University of Indonesia Indriyani , Adi; University of Indonesia Nugroho , Riri; University of Indonesia Fitri Sari , Dino; University of Indonesia Budi Prakoso
Dense long-range wide area network (LoRaWAN) deployments often experience reduced packet delivery reliability (PDR) and increased energy consumption due to suboptimal adaptive data rate (ADR) mechanisms. This study proposes a machine learning (ML) based joint optimization method to improve PDR or reliability while minimizing node energy consumption. The proposed solution employs random forest (RF) and light gradient boosting machine (LightGBM) models, which are suitable for resource constrained end devices (ED). Using combinations of data rate (DR), frequency channel, and transmission power (TxPower) from 500 ED, the model extracts 14 features for parameter optimization. Unlike conventional ADR, the proposed approach can learn complex relationships beyond the received signal strength indicator (RSSI) and signal-to-noise ratio (SNR) parameters. The model achieves higher prediction accuracy within a 5 km coverage area. Network simulator 3 (NS-3) simulation results demonstrate better performance and energy efficiency compared to conventional ADR. The proposed approach improves PDR by 6.1–11.6% and reduces node energy consumption by 20–25%, providing a better balance between communication reliability and energy efficiency in scalable LoRaWAN networks.
Volume: 24
Issue: 5
Page: 1581-1596
Publish at: 2026-10-01

NiteTron: a novel nighttime bunaken sea turtle species detection

10.12928/telkomnika.v24i5.27678
Muhamad; Sam Ratulangi University Dwisnanto Putro , Jane; Sam Ratulangi University Litouw , Abdul; Sam Ratulangi University Haris Ontowirjo , Sherwin Reinaldo; Sam Ratulangi University U. Sompie , Hebron; Sam Ratulangi University Prasetya
Automated monitoring is essential for sea turtle conservation, and vision-based algorithms offer a non-intrusive approach to observing turtles without disturb ing their natural habitat. However, nighttime conditions introduce significant challenges, as low illumination severely reduces the visibility of discriminative features. This limitation requires robust feature extraction to capture the lim ited important features. Recent advances in computer vision, such as you only look once (YOLO) 11-nano, enable real-time detection and allow optimization under challenging conditions. In this study, we propose NiteTron, a nighttime sea turtle species detector that uses a lightweight architecture with enhanced feature refinement. The network modifies YOLO11-nano to create a more ef ficient variant, YOLO11-pico. To improve detection in low-light scenarios, we introduce a strong feature modulation (SFM) module. This module enhances attention to relevant features by expanding the receptive field in the final back bone stage. We also incorporate a sparse separated convolution (S2C) module to extract key information efficiently while balancing accuracy and computational cost. The proposed model achieves high precision with fewer parameters and lower giga floating-point operations per second (GFLOPs) than YOLO11-nano. Experiments demonstrate superior performance, achieving 0.682 mean average precision (mAP)@0.5:0.95 for nighttime sea turtle detection. Inference tests also confirm efficient central processing unit (CPU) deployment, reaching 21.18 frames per second (FPS) on a desktop and 2.65 FPS on an edge device.
Volume: 24
Issue: 5
Page: 1561-1580
Publish at: 2026-10-01

Adaptive fuzzy proportional integral derivative control for azimuth rotational positioning system

10.12928/telkomnika.v24i5.27906
Mufid; UIN Sunan Gunung Djati Bandung Ridlo Effendi , Rina; UIN Sunan Gunung Djati Bandung Mardiati , Muhammad; UIN Sunan Gunung Djati Bandung Bagja Nasrulloh , Aan; Politeknik Manufaktur Bandung Eko Setiawan
This paper presents an adaptive fuzzy logic-based proportional integral derivative (PID) control strategy for an azimuth rotational positioning system characterized by nonlinear angular dynamics. Conventional fixed gain PID controllers often experience performance degradation when operating across varying error regions, resulting in slower responses and residual steady-state errors. To address this limitation, a region-based fuzzy gain scheduling mechanism is proposed to dynamically adjust the proportional (𝐾𝑝), integral (𝐾𝑖), and derivative (𝐾𝑑) gains according to the instantaneous angular error. The proposed controller is implemented on a laboratory-scale embedded platform consisting of a stepper motor actuator, potentiometer-based angular feedback, and an Arduino Nano controller. Experimental evaluations were conducted at reference angles of 45°, 90°, 135°, and 180° and compared with a conventional PID controller. Results demonstrate that the adaptive fuzzy–PID controller completely eliminates steady-state error at all tested setpoints and reduces rise time by up to 56.7%, depending on the operating condition. The adaptive gain adjustment enables faster transient response for large angular deviations while maintaining stable convergence near the setpoint. These findings demonstrate that the proposed embedded region-based adaptive fuzzy–PID approach provides an effective and practical solution for improving positioning accuracy and dynamic performance in rotational positioning systems.
Volume: 24
Issue: 5
Page: 1795-1804
Publish at: 2026-10-01

Robustness analysis of distributed CFAR detection with K-out-of-L fusion in heterogeneous environments

10.12928/telkomnika.v24i5.27784
Hung Van; Le Quy Don Technical University Pham , Tinh Dinh; Le Quy Don Technical University Nguyen
To operate effectively in complex environments, distributed radar systems re quire a robust constant false alarm rate (CFAR) processor. Conventional distributed cell-averaging (D-CA) CFAR detectors are optimal in homoge neous noise. However, they suffer severe performance degradation in non homogeneous scenarios, specifically under multiple interfering targets and abrupt clutter edges. This study presents a comprehensive performance analysis of distributed order statistics (D-OS) CFAR systems employing the K-out-of-L fusion rule. We derive the exact global detection and false alarm probabilities using the Poisson binomial distribution. These models are implemented via vari able precision arithmetic to ensure numerical stability under extreme masking conditions. Detection thresholds are rigorously calibrated to maintain a fixed global false alarm rate of 10−4 in homogeneous backgrounds. Quantitative re sults demonstrate that D-OS-CFAR significantly outperforms the D-CA-CFAR baseline. It maintains stable detection performance (PD > 0.9) under clutter power variation and suppresses false alarm surges at clutter edges to approxi mately 4.5 × 10−4. The analysis concludes that lower-K rules (e.g., K = 1,2) offer the optimal trade-off, limiting the homogeneous CFAR loss to below 0.5 dB while ensuring stable operation in highly heterogeneous environments.
Volume: 24
Issue: 5
Page: 1699-1709
Publish at: 2026-10-01

Analysis of software-defined network on Proxmox with quality of-service testing using Open vSwitch

10.12928/telkomnika.v24i5.27549
Shafira; STMIK AMIK Bandung Febriani , Riezkan; STMIK AMIK Bandung Aprianda Firmansyah
This study designs and implements a virtual network based on software defined networking (SDN) in the Proxmox Virtual Environment (VE) using Open vSwitch (OVS) and OpenDaylight (ODL) controller, with performance testing of quality of service (QoS). The parameters tested include rate limiter, throughput, packet loss, and forwarding rate. Proxmox VE is selected for its centralized virtualization and clustering capabilities, while OVS manages network traffic among virtual machines (VMs). ODL is chosen for its compatibility with OpenFlow and OVS database (OVSDB) protocols, essential for OVS management. Testing uses iPerf3 to generate network traffic and analyze performance, including database service access simulation to mimic typical cloud application traffic loads. Forwarding rate measures the system’s packet forwarding capability per unit time, providing a comprehensive network performance overview. Results contribute to optimized SDN-based virtual network implementations in Proxmox environments and provide scientific and practical references for developing QoS-based virtual networking systems.
Volume: 24
Issue: 5
Page: 1504-1512
Publish at: 2026-10-01

Engineering orchestration: adaptive and human-centred design for complexity, collaboration, and resilience

10.12928/telkomnika.v24i5.3777
Tole; Universitas Ahmad Dahlan Sutikno
Engineering design is undergoing a profound transformation as electrical and computer engineering systems become increasingly interconnected, intelligent, and embedded within dynamic socio-technical environments. Building upon systems thinking, future engineering can no longer rely solely on deterministic optimization or fixed design assumptions but must instead embrace adaptive, resilient, and human-centred approaches capable of responding to uncertainty and continuous change. This editorial argues that these developments fundamentally redefine engineering design as the orchestration of adaptive socio-technical systems that continuously align technological intelligence, human expertise, resilience, and lifecycle evolution. Adaptive engineering extends traditional design principles through continuous learning, feedback, and system reconfiguration, while human centred engineering ensures that technological capabilities remain aligned with human judgement, values, and operational needs. Resilience further enables engineering systems to anticipate, absorb, recover from, and adapt to evolving disturbances across increasingly complex operating environments. Engineering excellence therefore depends not merely on optimizing technological performance but on orchestrating these complementary capabilities into coherent engineering ecosystems capable of generating enduring value. This evolving perspective establishes a conceptual foundation for the future of electrical and computer engineering, where engineering orchestration enables adaptive design, human collaboration, resilient innovation, and ultimately responsible and sustainable engineering practice.
Volume: 24
Issue: 5
Page: 1449-1456
Publish at: 2026-10-01
1 2 3
...
2097 2098
Show 20 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