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

29,922 Article Results

Deep learning-based spam detection for WhatsApp chatbot fallback reduction

10.11591/ijai.v15.i1.pp909-918
Satrio Sadewo , Amalia Zahra
Chatbots on WhatsApp are widely used for customer service, but their effectiveness is often undermined by fallback responses when user input cannot be understood. A major cause of these fallbacks is unsolicited spam, which disrupts interactions and reduces service quality. This study develops and evaluates a spam detection system aimed at reducing fallback rates and enhancing user experience. A comparative analysis was conducted between traditional machine learning models (support vector machine (SVM) and decision tree (DT)) and advanced deep learning architectures, including long short-term memory (LSTM) variants (vanilla, bidirectional, stacked, convolutional neural network (CNN)-LSTM, and encoder-decoder) and transformer-based models (bidirectional encoder representations from transformers (BERT)-base, DistilBERT, and cross-lingual language model robustly optimized BERT pretraining approach (XLM-ROBERTa)). Using 170,000 messages sampled from 18 million interactions collected between July 2022 and December 2023, the models were assessed with standard evaluation metrics. Results show that CNN-LSTM and DistilBERT achieved the most robust performance. CNN-LSTM attained a precision of 0.92, recall of 0.91, F1-score of 0.91, and accuracy of 0.94, while DistilBERT achieved precision of 0.92, recall of 0.89, F1-score of 0.90, and accuracy of 0.93. These findings highlight their superior ability to capture contextual patterns in spam messages. Implementing such models is expected to significantly lower fallback rates, thereby improving chatbot reliability and user satisfaction.
Volume: 15
Issue: 1
Page: 909-918
Publish at: 2026-02-01

Brain tumor segmentation and classification using artificial hummingbird optimization algorithm

10.11591/ijai.v15.i1.pp429-442
Radhakrishnan Karthikeyan , Arappaleeswaran Muruganandham
The time and medical personnel experience are the only factors that determine whether brain tumors can be manually identified from numerous magnetic resonance imaging (MRI) pictures in medical practice. Many frameworks based on brain tumors are diagnosed using both deep learning and machine learning. This study proposes a Wasserstein deep convolutional generative adversarial network (WDCGAN) optimized using the artificial hummingbird optimization algorithm (AHBOA) for brain tumor segmentation and classification (SCBT). First, the BraTS dataset is used to gather the input data. Then it is pre-processed consuming adaptive self guided filtering (ASGF) and the result is segmented using fuzzy possibilistic C-ordered mean clustering (FPCOMC). After that, features are extracted using the dual tree complex discrete wavelet transform (DT-CDWT). The characteristics of feature extracted are fed to WDCGAN for effectively categorize the various parameters. Then the proposed MATLAB is used to implement the technique, and the performance measurements like F1-score, accuracy, error rate, precision, sensitivity, mean square error, receiver operating characteristic (ROC), and computational time are analyzed. The WDCGAN-AHBOA-SCBT method significantly improves precision in SCBT by integrating adaptive optimization strategies, resulting in 32.18, 32.75, and 32.90% higher precision in contrast to current techniques. This demonstrates that the approach is more accurate and effective, making it a reliable tool for medical diagnosis.
Volume: 15
Issue: 1
Page: 429-442
Publish at: 2026-02-01

Robust UAV localization of ground sensors in urban environments via path loss refinement and geometric selection

10.11591/ijai.v15.i1.pp412-428
Ahmed M. A. A. Elngar , Heng Siong Lim , Yee Kit Chan , Yaser Awadh Bakhuraisa , Ida Wahidah
Localizing ground sensors with unmanned aerial vehicles (UAVs) in dense urban environments is challenging because multipath and non-line-of-sight (NLoS) propagation distorts path loss (PL) measurements. This paper proposes a two-stage UAV localization framework that refines PL data and selects geometrically stable waypoint subsets before position estimation. In stage 1, PL samples are spatially smoothed by averaging measurements at neighboring UAV waypoints to reduce localized fluctuations. In stage 2, waypoint subsets are filtered using non-collinearity and non-adjacency constraints, and sensor positions are estimated using weighted least squares (WLS) and particle swarm optimization (PSO), with final estimates averaged across valid subsets. Wireless InSite ray-tracing simulations show that the framework reduces mean absolute error (MAE) from over 150 m to approximately 8.5 m. The proposed approach improves the practicality of UAV-assisted localization for urban internet of things (IoT) sensor deployments.
Volume: 15
Issue: 1
Page: 412-428
Publish at: 2026-02-01

Text summarization: BART, RF, and hybrid BART-RF algorithm comparison

10.11591/ijai.v15.i1.pp929-940
Muhammad Adib Zamzam , Agus Buono , Toto Haryanto
Data and information accumulate quantitatively and qualitatively. Abundant text data are posted on the internet. The number correlates to the complexity of the summarization. Automatic text summarization (ATS) is one of the most challenging tasks in natural language processing (NLP). ATS approached in three ways: extractive, abstractive, and hybrid. Hybrid approach combines both extractive and abstractive. This research tests and compares performance of bidirectional auto-regressive transformer (BART) and random forest (RF) individually and the performance combination of hybrid BART and RF in ATS. The research shows that individually, BART and RF recall-oriented understudy for gisting evaluation (ROUGE) scores are having quite differences. Consecutively, ROUGE RF scores in R1, R2, and RL are 51.45, 45.52, and 54.58 respectively. Meanwhile, ROUGE BART scores are 32.78, 16.17, and 32.19. Consecutively, average ROUGE RF, BART, and RF×BART F-measure are 45.73, 21.38, and 31.31. RF has the highest average score. ATS hybrid RF×BART is shown to be performed better than the default BART. The average ROUGE F-measures for RF×BART obtain moderate score at 31.31. This score is better than the default BART’s ROUGE score. RF×BART can be an alternative to the effective hybrid approach.
Volume: 15
Issue: 1
Page: 929-940
Publish at: 2026-02-01

Centrality-optimized coalition formation: a genetic algorithm approach with leadership attributes

10.11591/ijai.v15.i1.pp383-398
Anon Sukstrienwong , Sorapak Pukdesree
In graph theory, centrality is often assessed using traditional methods such as closeness centrality, which measures the average shortest path length between nodes in a network. In this study, we primarily focus on developing the proposed approach and demonstrating its effectiveness through initial experimental results. A novel genetic algorithm (GA)–based method named centrality–optimized leadership coalition formation (COLCF) has been designed. It emphasizes actual agent distances according to closeness centrality and leadership attributes in group formation. We detail the COLCF algorithm, present empirical case studies, and provide efficiency comparisons. In accordance with our simulation results, the proposed algorithm is capable of capitalizing on the ideal coalition structure for achieving high closeness centrality when incorporated with leadership attributes. The experimental results demonstrate the algorithm’s robustness and effectiveness in addressing complex coalition formation challenges.
Volume: 15
Issue: 1
Page: 383-398
Publish at: 2026-02-01

Design of Antasena: an AI-powered maritime surveillance and anomaly detection system for security decision support

10.11591/ijai.v15.i1.pp269-288
Arif Badrudin , Siswo Hadi Sumantri , Rudy Agus Gemilang Gultom , I Nengah Putra Apriyanto , Umi Laili Yuhana , Fitria Dwi Ratnasari
Indonesia’s vast maritime territory faces serious challenges from illegal fishing, smuggling, and habitat destruction. To address these, the Indonesian Navy (TNI-AL) developed Antasena, an artificial intelligence (AI)-powered smart dashboard integrating automatic identification system (AIS) data, satellite imagery, and conservation metrics. Antasena leverages advanced anomaly detection algorithms, achieving 95.3% accuracy, 94.7% precision, 94.2% recall, and a 96.8% receiver operating characteristic-area under the curve (ROC-AUC) score in identifying vessel anomalies, including unauthorized fishing and smuggling activities. Using the analyze, design, develop, implement, and evaluate (ADDIE) framework, the system supports real-time maritime surveillance and biodiversity monitoring in conservation zones. The main contributions of this study include the development of a user-centric AI-based dashboard for maritime anomaly detection, the integration of multi-source data with machine learning models, and validation through operational field tests with maritime authorities. Antasena offers a scalable and effective solution to strengthen maritime security and protect Indonesia’s marine resources.
Volume: 15
Issue: 1
Page: 269-288
Publish at: 2026-02-01

Neuro-DANet: dual attention deep neural network long short term memory for autism spectrum disorder detection

10.11591/ijai.v15.i1.pp810-823
Sujatha Hanumantharayappa , Manjula Rudragouda Bharamagoudra
Autism spectrum disorder (ASD) is neurological illness affects ability of individuals to communicate and interact socially, and it is diagnosed in any time. Early detection of ASD is especially significant due to its subtle characteristics and high costs associated with the detection process. Traditional deep learning (DL) models struggle to capture intricate spatiotemporal dependencies in functional magnetic resonance imaging (fMRI) data, resulting in minimized detection performance and poor generalization. To address these drawbacks, the proposed Neuro-DANet combines a dual-attention deep neural network (DA-DNN) with long short term memory (LSTM) to efficiently learn spatial and temporal features from fMRI scans. The continuous wavelet transform (CWT) is used to extract multi-scale features and the principal component analysis (PCA) is utilized to dimensionality reduction, which enhances robustness and efficacy. The dual self-attention mechanism improves the interpretability of the model by focusing on critical brain regions and time steps that are most relevant to ASD severity. The developed Neuro-DANet obtains the highest accuracy of 98.51% on autism brain imaging data exchange (ABIDE)-I and 98.81% on ABIDE-II datasets when compared with traditional algorithms.
Volume: 15
Issue: 1
Page: 810-823
Publish at: 2026-02-01

Artificial intelligence in orthodontics: modeling decision support systems for treatment planning

10.11591/ijai.v15.i1.pp97-105
Sowmya Lakshmi Belur Subramanya , Advaith Vijaya Mohan , Achala Varsha Vishlavath Premalatha , Manchikanti Varunsai
Orthodontic treatment planning involves complex clinical decision-making that can benefit from artificial intelligence (AI). This study evaluates machine learning and deep learning models—including random forest, AdaBoost, gradient boosting, and artificial neural networks (ANNs)—for predicting orthodontic treatment strategies using a dataset of 612 anonymized patient records with 66 clinically validated features across four categories (extraction, non-extraction, functional appliance, and orthopedic case). Preprocessing included imputation, normalization, and the synthetic minority oversampling technique (SMOTE) for class imbalance, while performance was assessed via 10-fold cross-validation. Results showed that ANNs achieved the highest balanced accuracy (0.83), F1-score (0.84), and receiver operating characteristic area under the curve (ROC-AUC) (0.90), outperforming ensemble and baseline models. Shapley additive explanations (SHAP) analysis confirmed clinically meaningful predictors such as vertical face proportions and mandibular plane angle, enhancing interpretability. Although promising, the study is limited by its single-institution dataset and lack of external validation. Future research should incorporate multicenter, multimodal datasets and interpretable-by-design frameworks to enable clinically trusted AI decision-support systems in orthodontics.
Volume: 15
Issue: 1
Page: 97-105
Publish at: 2026-02-01

Automated data exploration with mutual information in natural language to visualization

10.11591/ijai.v15.i1.pp129-139
Hue Thi-Minh Luong , Vinh-The Nguyen , Van-Viet Nguyen , Kim-Son Nguyen , Huu-Khanh Nguyen
Transcribing natural language to visualization (NL2VIS) has been investigated for years but still suffer from several fundamental limitations (e.g., feature selection). Although large language models (LLMs) are good candidates but they incur computation cost and hard to trace their made decisions. To alleviate this problem, we introduced an alternative information-theoretic framework that utilized mutual information (MI) to quantify the statistical relationship between utterances and database features. In our approach, kernel density estimation (KDE) and neural estimation techniques were utilized to estimate MI, and to optimize a diversity-promoting objective balancing feature relevance and redundancy. We also introduced the information coverage ratio (ICR) to quantify the amount of information content preserved in feature selection decisions. In our experiments, we found that the proposed approach improved information-theoretic metrics, with F1-score of 0.863 and an ICR of 0.891. We observed that these improvements did not come at the cost of traditional benchmarks: validity reached 88.9%, legality 85.2%, and chart-type accuracy 87.6%. Moreover, significance tests (p < 0.001) and large effect sizes (Cohen’s d > 0.8) further supported that these improvements were meaningful for feature selection. Thus, this study provides a mathematical framework for applications requiring analytical validity that extends beyond NL2VIS to other machine learning contexts.
Volume: 15
Issue: 1
Page: 129-139
Publish at: 2026-02-01

Adaptive deformable feature augmentation and refinement network for scene text detection and recognition

10.11591/ijai.v15.i1.pp831-840
Ratnamala S. Patil , Geeta Hanji , Rakesh Hudud
Scene text recognition (STR) is the task of detecting and identifying text within images captured from natural scenes, a challenging process due to variations in text appearance, orientation, and background complexity. The proposed methodology, adaptive deformable feature augmentation and refinement network (ADFARN), is designed to address these challenges by combining deformable convolutional networks for robust enhanced feature extraction with a novel deep feature refinement (FRE) that leverages refinement for precise text localization. This approach enhances the differentiation between text and background, significantly improving recognition accuracy. The ADFARN methodology includes a comprehensive process of feature extraction, deep feature augmentation module (DFAM), and the generation of score and threshold maps through differentiable binarization. The adaptive nature of the model allows it to handle low resolution and partially occluded text effectively, further increasing its robustness. Additionally, the proposed method aligns visual and textual features seamlessly. Extensive performance evaluation on the common objects in context (COCO)-Text dataset demonstrates that ADFARN outperforms existing state-of-the-art methods in terms of precision, recall, and F1-scores, establishing it as a highly effective solution for STR in real world applications.
Volume: 15
Issue: 1
Page: 831-840
Publish at: 2026-02-01

A survey on leveraging artificial intelligence tools for enhancing advanced mathematical education and problem-solving

10.11591/ijai.v15.i1.pp76-85
Hadeel N. Abosaooda , Syaiba Balqish Ariffin , Osamah Mohammed Alyasiri
Artificial intelligence (AI) has increasingly shaped education, with ChatGPT developed by OpenAI, emerging as a prominent tool due to its ability to generate contextually relevant language and support learning. This survey investigates the integration of ChatGPT into mathematics education, focusing on three dimensions. First, it explores innovative strategies for creating interactive and personalized learning environments that adapt to individual student needs. Second, it evaluates ChatGPT’s specific advantages in mathematics instruction, including providing tailored feedback, assisting with problem-solving, and deepening conceptual understanding. Third, it addresses the challenges of adopting ChatGPT in advanced mathematics education, such as risks of over-reliance, the necessity of balancing AI with traditional pedagogy, and the importance of ongoing professional development for educators. Recent studies highlight ChatGPT’s potential to solve complex mathematical problems, such as those in linear algebra and word problems, while also noting limitations related to accuracy and the preservation of critical thinking skills. The findings demonstrate that ChatGPT can significantly enhance mathematics education by supporting personalized learning and complex problem-solving. Therefore, this study will contribute to the discourse on AI in education by identifying opportunities, challenges, and implications for equity, pedagogy, and the responsible integration of ChatGPT in future classrooms.
Volume: 15
Issue: 1
Page: 76-85
Publish at: 2026-02-01

AI-powered hub optimization: a reinforcement learning and graph-based approach to scalable blockchain networks

10.11591/ijai.v15.i1.pp536-546
Kassem Danach , Hassan Rkein , Alaaeddine Ramadan , Hassan Harb , Bassam Hamdar
Blockchain networks face persistent scalability challenges, including network congestion, high latency, and transaction costs. To address these limitations, this study proposes an AI-driven hub location optimization framework that integrates reinforcement learning (RL), mixed integer linear programming (MILP), and graph neural networks (GNNs). The RL-based hub selection dynamically identifies optimal supernode placement, while MILP ensures cost-efficient transaction routing, and GNNs predict flow patterns for proactive congestion management. Experimental results on Ethereum and Bitcoin datasets demonstrate significant improvements, including a 58.6% reduction in transaction latency, 28.9% gas fee savings, and 41.5% congestion reduction. Beyond performance gains, statistical tests confirm the significance of these improvements, and ablation studies highlight the complementary role of each component.
Volume: 15
Issue: 1
Page: 536-546
Publish at: 2026-02-01

Hybrid texture-deep feature fusion for mammogram classification: a patient-level, calibrated evaluation

10.11591/ijai.v15.i1.pp861-877
Muhammad Subali , Lulu Mawadddah Wisudawati , Teresa Teresa
We propose a lightweight computer-aided diagnosis (CAD) framework that fuses four sub-band discrete wavelet transform gray-level co-occurrence matrix (DWT–GLCM) texture features with fine-tuned ResNet-50 embeddings under a strict, patient-level, leak-free evaluation protocol. Experiments were conducted on two public datasets: mammographic image analysis society (MIAS) (normal vs. abnormal) and curated breast imaging subset of the digital database for screening mammography (CBIS-DDSM) (benign vs. malignant). Five-fold cross-validation (CV) was confined to the training portion, operating thresholds were fixed on the validation split to target high recall, and the held-out test set was evaluated once. Performance was assessed using accuracy, F1-score, receiver operating characteristic (ROC)-area under the curve (AUC) with bootstrap 95% confidence intervals (CI), precision-recall (PR)-AUC, and calibration metrics (Brier score, expected calibration error). The proposed fusion model achieved ROC-AUC on MIAS (0.992) and strong performance on CBIS-DDSM (0.896), with consistent PR characteristics. Calibration analysis indicated reliable probability estimates and clinically interpretable decisions at a 95% sensitivity operating point. Ablation experiments revealed substantial gains over texture-only baselines and parity with convolutional neural network (CNN)-only models, highlighting fusion as a simple yet well-calibrated alternative for screening-oriented workflows. This study underscores the necessity of patient-level evaluation, explicit operating-point selection, and calibration reporting to ensure clinically meaningful CAD performance in mammography.
Volume: 15
Issue: 1
Page: 861-877
Publish at: 2026-02-01

TAHRF: enhancing personalized tourism recommendations with dynamic adaptation

10.11591/ijai.v15.i1.pp374-382
Mohamed Badouch , Mehdi Boutaounte
The rapid growth of online tourism data intensifies information overload, while conventional recommender systems struggle with sparsity, cold-start issues, and single-criteria ratings. This paper presents the trust-aware hybrid recommendation framework (TAHRF), which integrates user-item trust propagation, multi-criteria ratings, and dynamic preference adaptation. TAHRF employs Euclidean-Jaccard trust metrics, item connectivity, and rating consistency, combined with a feedback-driven weighting mechanism. Experiments on TripAdvisor datasets show superior performance: mean absolute error (MAE) reduced to 0.98 (restaurants) and 0.71 (hotels), outperforming multi-criteria tensor-based collaborative filtering (MC-TeCF) baselines. TAHRF also achieves higher precision@5, with coverage maintained under extreme sparsity. Ablation studies confirm the critical role of trust propagation, multi-criteria analysis, and adaptive weighting. TAHRF advances personalized, transparent, and adaptive tourism recommendations.
Volume: 15
Issue: 1
Page: 374-382
Publish at: 2026-02-01

Explainable deep learning for scalable record linkage: a TabNet-based framework for structured data integration

10.11591/ijai.v15.i1.pp725-743
Fatima Zahrae Saber , Ali Choukri , Mohamed Amnai , Abderrahim Waga
Record linkage is considered a fundamental process for ensuring data quality and reliability, with critical applications in domains such as healthcare, finance, and commerce. A machine learning-based approach for optimizing record linkage in structured datasets is presented in this paper. By integrating hybrid blocking methods (combining standard blocking and sorted neighborhood approaches) with advanced similarity measures, computational overhead is significantly reduced while high accuracy is maintained. The performance of TabNet, a deep learning model designed for tabular data, is compared with traditional deep neural networks (DNNs) in the classification phase. Experimental results on a synthetic dataset of 5,000 records demonstrate that comparable precision and recall are achieved by TabNet to DNNs while execution time is reduced by over 79%. The scalability and efficiency of the proposed method are highlighted by these findings, making it well-suited for large-scale data management tasks. Practical and computationally efficient solutions for record linkage in the era of big data are contributed to by this work.
Volume: 15
Issue: 1
Page: 725-743
Publish at: 2026-02-01
1 2 3
...
1994 1995
Show 40 of 1995

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