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

Intelligent plant disease detection using twin attention optimal convolutional neural network

10.11591/ijai.v15.i1.pp756-765
Prameetha Pai , Namitha S. J. , Sowmya T. , Amutha S. , Nisarga Gondi
Farming is one of the most important ways for people in India to make a living. Rice is a staple food, and when farmers successfully harvest rice crops, pests often attack them, which costs agriculture a lot of money. There are now a lot of new AI-based ways to help with this problem in rice plants. But those ways don’t work very well because they take a long time and make mistakes when sorting things. This article talks about a new hybrid deep learning (DL) method for finding leaf diseases in rice plants. This process has four main steps: pre-processing, segmentation, feature extraction, and classification. A hybrid DL-based twin attention convolutional neural network (CNN) model classifies segmented images into healthy and unhealthy leaves. But this method has the problem of overfitting. An optimization method based on chaotic slime mould (CSM) solves this problem. The proposed method is compared with bidirectional long short-term memory (Bi-LSTM), recurrent neural network (RNN), deep neural network (DNN), and deep belief network (DBN). The suggested method has an overall accuracy of 99.56%, an F-measure of 99.21%, a sensitivity of 99.16%, a specificity of 98.56%, a precision of 99.26%, a mean absolute error (MAE) of 0.004, a mean squared error (MSE) of 0.004, and a root mean square error (RMSE) of 0.06.
Volume: 15
Issue: 1
Page: 756-765
Publish at: 2026-02-01

Dynamic attack pattern-aware intelligent cyber-physical intrusion detection system for internet of things-edge networks

10.11591/ijai.v15.i1.pp580-591
Vishala Ibasapura Lakshminarayanappa , Kempahanumaiah M. Ravikumar
The proliferation of Internet of Things (IoT) technologies, coupled with the convergence of edge computing infrastructures, has revolutionized modern cyber-physical systems (CPS). However, the inherently distributed architecture of these systems increases their vulnerability to advanced network-level cyber threats, posing significant challenges to data integrity and system reliability. Traditional machine learning (ML) and deep learning (DL)-based intrusion detection systems (IDS) often fall short in identifying evolving attack vectors due to their limited adaptability. To address these limitations, this paper introduces a novel Dynamic Attack Pattern-Aware Improvised Weighted Gradient Boosting (DAPA-IWGB) model designed to enhance real-time threat detection and adaptive response within IoT-edge-enabled CPS environments. The DAPA-IWGB framework synergizes gradient tree boosting with an enhanced loss function handling covariate shift, while incorporating statistical monitoring mechanisms for dynamic covariate shift recognition and continuous learning. Comprehensive experimental validation using two prominent benchmark datasets ToN-IoT and UNSW-NB15 demonstrates the proposed model’s robustness and superior performance, achieving detection accuracies of 99.921% and 99.93%, respectively. Comparative evaluations highlight substantial improvements in detection accuracy, adaptability, and reliability over existing IDS solutions. The results affirm the effectiveness of the DAPA-IWGB model in fortifying the security posture of distributed IoT-based CPS against sophisticated and evolving cyber threats.
Volume: 15
Issue: 1
Page: 580-591
Publish at: 2026-02-01

Efficient YOLO-based models for real-time ceramic crack detection

10.11591/ijai.v15.i1.pp852-860
Benchalak Maungmeesri , Sasithorn Khonthon , Dechrit Maneetham , Padma Nyoman Crisnapati
The following research work systematically compares four variants of you only look once (YOLO), namely, YOLOv8, YOLOv9, YOLOv10, and YOLOv11 proposed recently, considering the key properties required to perform ceramic surface crack detection tasks with high computational efficiency, real-time inference speed, and low memory usage. A total of 300 images of ceramic surfaces were collected with manually labeled cracks and divided into training, validation, and testing sets in portions of 263, 22, and 15 images, respectively. Each of the four YOLO variants was trained for 50 and 100 epochs, and each was evaluated regarding mean average precision (mAP), inference time, model size, and computational complexity in giga floating point operations per second (GFLOPs). YOLOv9 produced the highest accuracy with mAP values as high as 0.752-0.79 but the highest cost in terms of increased computational complexity. However, among these methods, YOLOv8 can produce the fastest inference (~2-2.3 ms) with a small memory footprint (~6 MB) with an acceptable accuracy of ~0.65-0.67. The results showed that YOLOv8 is the most feasible to deploy in resource constrained industrial automation environments. By offering this comparative study, the research attempts to provide hints for the selection of appropriate YOLO-based models by practitioners in quality control applications related to ceramic manufacturing.
Volume: 15
Issue: 1
Page: 852-860
Publish at: 2026-02-01

Classifying mental workload of esports players using machine learning

10.11591/ijai.v15.i1.pp469-480
Aisy Al Fawwaz , Osmalina Rahma , Sayyidul Istighfar Ittaqillah , Angeline Shane Kurniawan , Revita Novianti Putri , Richa Varyan , Aura Adinda , Khusnul Ain , Rifai Chai
Electrodermal activity (EDA) peak counts, derived from both tonic and phasic components, are widely used as physiological proxies for mental workload in cognitively demanding tasks, such as esports. However, their specificity remains uncertain, particularly given potential confounding effect of time-on-task. This study analyzes 92 competitive gameplay sessions from a multimodal esports dataset using three decomposition techniques: convex decomposition (cvxEDA), sparse deconvolution (sparseEDA), and time varying sympathetic activity (TVSymp). From each method, phasic, and tonic peak counts (TPC), as well as their normalized rates, were extracted. We examined their relationship with self-reported workload through correlation analyses, partial correlations controlling for session duration, and linear mixed-effects models (LMMs). While both peak types exhibited strong positive correlations with gameplay duration (r=0.915 for phasic and r=0.856 for tonic), their association with perceived workload vanished once time was accounted for. Across methods, TVSymp yielded the highest discriminative validity with an area under curve (AUC) of 0.880 in classifying high versus low workload. Machine learning (ML) classifiers trained solely on EDA-based features under a leave-one-subject-out (LOSO) scheme outperformed multimodal models that incorporated heart rate variability (HRV). These results underscore need to disentangle temporal structure from cognitive signals when interpreting EDA and call into question the assumption that EDA peak counts alone reliably encode mental workload across individuals.
Volume: 15
Issue: 1
Page: 469-480
Publish at: 2026-02-01

Hybrid convolutional networks, hidden Markov models, and autoencoders for enhanced recognition

10.11591/ijai.v15.i1.pp780-787
Driss Naji , Kamal Elhattab , Abdelali Joumad , Abdelouahed Ait Ider , Abdelkbir Ouisaadane , Azzeddine Idhmad
Recognition problems, including object detection, scene understanding, and fine-grained categorisation, are popular subjects in computer vision. However, it is challenging to model spatial coherence and contextual dependencies in response to changes in configurations. Human Vs computers' ability in perception-although convolutional neural networks (CNNs) do well in the extraction of features, they have high dependence on local receptive fields and are not able to capture long-range spatial relationships and high-order interactions. To alleviate the shortcomings of the current approaches, we present an enhanced hybrid CNNs two dimensional hidden Markov model (2D-HMM) framework that combines 2D-HMM, Markov random fields (MRF) and variational autoencoders (VAEs) into a single model. The model employs 2D-HMMs for pairwise spatial modelling, MRFs for higher order context, and VAEs for stable latent representation learning. Tested on the MNIST and CIFAR-10 benchmark datasets, our approach consistently outperforms the state-of-the-art performance by 98.2% and 89.5%, respectively, with high robustness to noise and occlusion. Results from ablation studies further show that MRFs improve recall by 1.6% and VAEs improve precision by 1.3%, suggesting that they complement each other sufficiently with respect to overall testing performance. This work unifies deep learning and probabilistic graphical models, leading to more interpretable, scalable, and accurate recognition systems.
Volume: 15
Issue: 1
Page: 780-787
Publish at: 2026-02-01

Review of artificial intelligence in smart wearable devices under internet of things communication

10.11591/ijai.v15.i1.pp1-11
Minh Long Hoang , Guido Matrella , Paolo Ciampolini
This paper aims to provide a review about the role of artificial intelligence (AI) in wearable devices, specifically smartwatches, fitness trackers, smart clothes, and smart eyewear. Machine learning (ML) and deep learning (DL) play essential roles in the development of these devices, thanks to their advanced algorithms with the support of the internet of things (IoT) framework. AI functionalities and metrology are detailed in these wearables, highlighting the use of convolutional neural networks (CNN) and recurrent neural networks (RNN) for applications such as activity recognition, health monitoring, and personalized recommendations. The paper demonstrates the AI implementation in smart devices, including stress detection by heart rate variability (HRV), personalizing fitness recommendations, muscle activity monitoring, and real-time image recognition. Challenges and potential solutions are discussed for a deep comprehension of the AI development in wearable devices.
Volume: 15
Issue: 1
Page: 1-11
Publish at: 2026-02-01

Deformable spatial pyramid pooling-enhanced EfficientNet with weighted feature fusion for pomegranate fruit disease diagnosis

10.11591/ijai.v15.i1.pp642-654
Harish Bommenahalli Mallikarjunaiah , Balaji Prabhu Baluvaneralu Veeranna
Pomegranate is a fruit of high nutritional and economic importance. Still, it is highly susceptible to different diseases during its growing stages, leading to significant yield losses and financial setbacks for farmers. This article proposes a novel disease detection model that integrates handcrafted features with deep features extracted using a developed deformable spatial pyramid pooling (DSPP)-EfficientNet architecture. Handcrafted features such as color (RGB and HSV histograms), texture features from gray level co occurrence matrix (GLCM), and shape attributes extracted from contour descriptors and Hu moments are captured and fused with deep features by weighted fusion strategy, resulted in the most discriminative information. The fused features are categorized using a support vector machine (SVM) in a classification phase, which effectively classifies different classes of pomegranate fruit diseases. The combined deep and handcrafted features obtained 96.66% accuracy, 96.26% precision, 96.50% recall, 96.37% F1 score, and 95.64% specificity on the pomegranate fruit disease dataset which compared to existing techniques.
Volume: 15
Issue: 1
Page: 642-654
Publish at: 2026-02-01

Exploring cutting-edge research, applications, and future directions in artificial intelligence across diverse domains

10.11591/ijai.v15.i1.pp1019-1022
Tole Sutikno
This issue highlights the most recent advances in artificial intelligence (AI) research, which cover a wide range of applications, methodologies, and emerging technologies. The gathered works highlight AI's transformative potential in a variety of fields, including healthcare, environmental monitoring, energy systems, intelligent transportation, cybersecurity, smart agriculture, and human-computer interactions. The featured studies demonstrate novel applications of deep learning, convolutional neural networks, vision transformers, reinforcement learning, ensemble methods, and explainable AI techniques, with a focus on both performance optimisation and interpretability. The issue also delves into AI integration with IoT, blockchain, big data, and mobile platforms, showcasing scalable and real-time solutions for dynamic, data-intensive settings. Aside from technical accomplishments, the contributions address practical issues such as model generalisation, feature selection, data quality, privacy, and ethical concerns. These works show how AI is improving decision-making, predictive capabilities, and operational efficiency while addressing complex societal and industrial issues. Looking ahead, this issue encourages reflection on AI development trajectories, emphasising the importance of robust, explainable, and adaptive systems that balance computational power and interpretability. This issue aims to inform, inspire, and guide practitioners and researchers in shaping the next generation of intelligent technologies by providing a comprehensive overview of cutting-edge AI research and real-world applications.
Volume: 15
Issue: 1
Page: 1019-1022
Publish at: 2026-02-01

Real-time intelligent virtual assistant based on retrieval augmented generation

10.11591/ijai.v15.i1.pp237-246
I Ketut Resika Arthana , Ni Putu Novita Puspa Dewi , Gede Arna Jude Saskara , I Made Ardwi Pradnyana , Luh Indrayani
Improving user experience in accessing information on organizational websites remains a challenge. Users often face complex navigation and multi step searches that slow information retrieval. This study introduces the real time intelligent virtual assistant (RIVA), which integrates large language models (LLMs) with the retrieval-augmented generation (RAG) framework to support real-time interaction with website content. The system was implemented on the Universitas Pendidikan Ganesha (Undiksha) website using a WordPress content management system (CMS) and developed following the design science research (DSR) approach, which includes six stages: problem identification, solution objectives, design and development, demonstration, evaluation, and communication. The retrieval-augmented generation assessment (RAGAS) evaluation indicated that the combined model of text-embedding-ada-002 and semantic chunking yielded the best results, with context precision=0.83, context recall=0.90, response relevancy=0.91, faithfulness=0.83, and answer correctness=0.85. User experience questionnaire (UEQ) testing performed well, particularly in the novelty and stimulation dimensions. These results demonstrate that RIVA can provide users with access to relevant and engaging information. As a result, future research will focus on improving retrieval and developing adaptive semantic chunking for structured and complex data.
Volume: 15
Issue: 1
Page: 237-246
Publish at: 2026-02-01

Review of ChatGPT tools in education systems based on literature

10.11591/ijai.v15.i1.pp12-19
Siva Prasad Reddy K. V. , Parvathi Malepati , Khadarbadar Pullamma , Gangapuram Mallikarjunachari , Sareddy Venkata Rami Reddy , Anil Kumar Vadaga , Sai Shashank Gudla
Artificial intelligence (AI) has rapidly reshaped modern education, with ChatGPT emerging as one of the most influential generative AI tools supporting teaching, learning, and academic administration. This review synthesizes evidence from 65 peer-reviewed studies published since 2022 to evaluate ChatGPT’s educational applications, benefits, constraints, and ethical implications. Findings indicate that ChatGPT enhances personalized learning, academic writing, digital literacy, and instructional efficiency, while offering scalable support for large classrooms. Comparative analyses reveal that ChatGPT demonstrates superior linguistic coherence and reasoning compared to Gemini, Bing Chat/Copilot, and Claude. However, concerns persist regarding hallucinations, academic dishonesty, data privacy, infrastructural disparities, and faculty readiness. The review highlights the need for responsible governance frameworks, AI literacy programs, and equitable institutional policies. Future directions include longitudinal research on learning outcomes, inclusive AI design, cross-cultural adoption patterns, and evolving teacher–student dynamics in AI-augmented environments.
Volume: 15
Issue: 1
Page: 12-19
Publish at: 2026-02-01

Web-based geothermal drilling stuck pipe prediction using decision tree algorithm

10.11591/ijai.v15.i1.pp604-614
Rosyihan Muhtadlor , Nur Rohman Rosyid , Anni Karimatul Fauziyyah , Lalu Hendra Permana Setiawan , Irfan Saputra , Pavel Stasa , Filip Benes , Muhammad Syafrudin , Ganjar Alfian
In geothermal drilling operations, data from rig-mounted sensors play a crucial role in maintaining operational efficiency and preventing drilling failures. However, sensor uncertainties and complex subsurface conditions can lead to stuck pipe incidents, causing significant non-productive time and financial losses. This study proposes web-based drilling monitoring system integrated with machine learning (ML) to predict stuck pipe occurrences in geothermal drilling. Several ML algorithms—decision tree (DT), random forest (RF), naïve Bayes (NB), multilayer perceptron (MLP), and support vector machine (SVM)—were evaluated using geothermal drilling data from an Indonesian geothermal project conducted in 2023. To address class imbalance, the synthetic minority oversampling technique (SMOTE) was applied to the training dataset. Feature selection was performed using the correlation coefficient method, and predictions were generated using a 5 minute sliding window. Among the evaluated models, the DT consistently demonstrated superior performance across multiple prediction horizons (PH), achieving an accuracy of 97.4%, precision of 98.6%, recall of 72.9%, and a ROC-AUC of 0.729 using the top five selected features. The trained model was integrated into web-based monitoring platform that provides visualization and predictive alerts. This system enables early detection and better decision-making, helping improve drilling efficiency, reduce stuck pipe risks, and enhance operational safety.
Volume: 15
Issue: 1
Page: 604-614
Publish at: 2026-02-01

Evolutionary trends in automatic speech recognition with artificial intelligence: a systematic literature review

10.11591/ijai.v15.i1.pp20-43
Gabriel Oluwatobi Sobola , Emmanuel Adetiba , Olabode Idowu-Bismark , Abdultaofeek Abayomi , Raymond Jules Kala , Surendra Colin Thakur , Sibusiso Moyo
Human beings depend greatly on communication and continually seek ways to overcome language barriers. Automatic speech recognition (ASR) has emerged as a vital tool for enhancing human interaction. Early ASR research relied on probabilistic models, particularly the hidden Markov model (HMM) and Gaussian mixture model (GMM), with mel-frequency cepstral coefficients (MFCCs) for feature extraction, leading to the creation of Audrey at Bell Laboratories. Subsequently, artificial intelligence (AI) approaches, especially deep learning, have transformed ASR and produced systems such as Jasper, Whisper, Google Assistant, Microsoft Cortana, Apple Siri, and Amazon Alexa. This paper presents a systematic literature review that examines ASR’s evolution, the AI architectures employed, their features, strengths and weaknesses, and the performance gains achieved since AI was integrated into probabilistic modelling. A snowballing approach was used to identify relevant studies from Google Scholar and Scopus to address five research questions, iterating through backward and forward searches until no new information was found. Findings reveal that ASR dates back to the 1920s with the Radio Rex toy and has since advanced through architectures including deep learning, recurrent neural networks (RNN), support vector machines (SVM), and transformers, all contributing to improved performance measured by reduced word error rates (WER).
Volume: 15
Issue: 1
Page: 20-43
Publish at: 2026-02-01

Machine learning models in the enhancement of PSE in high-dimensional socioeconomic data: a review

10.11591/ijeecs.v41.i2.pp645-654
Gene Marck B. Catedrilla , Joey Aviles
This study reviews the use of machine learning (ML) techniques to improve propensity score (PS) estimation in high-dimensional socioeconomic data. Traditional logistic regression (LR) often performs poorly under nonlinear and complex covariate structures, leading to bias and model misspecification. Across the reviewed studies, ensemble methods such as random forests (RF) and gradient boosting, and deep learning models consistently achieved better covariate balance, lower bias, and greater flexibility than conventional approaches, while classification-based methods improved performance in imbalanced datasets. The review also highlights practical considerations, including calibration, transparent reporting, and integration with doubly robust estimators to strengthen causal inference. The findings show that ML-based propensity score estimation (PSE) can substantially enhance the validity and reliability of socioeconomic evaluations, provided that its implementation is carefully guided by appropriate expertise and best-practice standards.
Volume: 41
Issue: 2
Page: 645-654
Publish at: 2026-02-01

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

A novel BERT-long short-term memory hybrid model for effective credit card fraud detection

10.11591/ijai.v15.i1.pp788-797
Oussama Ndama , Safae Ndama , Ismail Bensassi , El Mokhtar En-Naimi
In the rapidly evolving landscape of financial transactions, the detection of fraudulent activities remains a critical challenge for financial institutions worldwide. This study introduces a novel bidirectional encoder representation from transformers (BERT)–long short-term memory (LSTM) hybrid model that integrates both textual and numerical data to enhance credit card fraud detection. Leveraging BERT for deep contextual embeddings and LSTM for sequence analysis, the model provides a comprehensive approach that surpasses traditional fraud detection systems primarily based on numerical analysis. On the validation set, the model achieved a recall of 100% and an accuracy of 99.11%, highlighting strong effectiveness in identifying fraudulent transactions under class imbalance. Through rigorous evaluation, the model demonstrated exceptional accuracy and reliability, promising improvements in fraud detection and mitigation. This paper details the development and validation of the hybrid model, emphasizing its use of mixed data types to capture complex patterns in transaction data. The results indicate a new frontier in fraud detection by combining natural language processing (NLP) and sequential data analysis to create a robust solution for real-world applications, supporting the security and integrity of financial systems globally.
Volume: 15
Issue: 1
Page: 788-797
Publish at: 2026-02-01
1 2 3
...
1994 1995
Show 39 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