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

Spark-powered bioactivity prediction: a comparison of machine learning approaches

10.11591/ijai.v15.i3.pp2423-2430
Nazif Tchagafo , Abderrahmane Ez-Zahout , Ahiod Belaid
The arduous and expensive nature of drug discovery has long been a bottleneck in scientific progress. However, recent breakthroughs in computational power, notably machine learning (ML) and artificial intelligence (AI), are profoundly transforming the field. Automated machine learning (AutoML) presents itself as a significant advancement, streamlining model selection, and hyperparameter tuning. This study delves into the potential of AutoML to accelerate drug discovery by comparing it to classical ML techniques. The focus lies on predicting the bioactivity of epidermal growth factor receptor (EGFR), a critical protein implicated in many cancers. By utilizing the scalability of Apache Spark, vast and diverse datasets encompassing biological, chemical, and genomic data tied to EGFR are processed. This comparative analysis aims to evaluate the comparative performance of both approaches, thereby contributing actionable insights to drug discovery research.
Volume: 15
Issue: 3
Page: 2423-2430
Publish at: 2026-06-01

Enhancing phishing website detection: a comparative study of SMOTETomek-XGB and SMOTEENN-XGB

10.11591/ijai.v15.i3.pp2935-2945
Kamal Omari , Ayoub Oukhatar
In the evolving landscape of cybersecurity, phishing websites continue to be a persistent threat, challenging detection methods due to the significant class imbalance between phishing and legitimate websites. This study evaluates the effectiveness of two advanced hybrid-resampling techniques SMOTETomek and SMOTEENN integrated with the extreme gradient boosting (XGBoost) classifier to enhance phishing website detection. SMOTETomek combines the synthetic minority over-sampling technique (SMOTE) with Tomek links, creating synthetic examples and eliminating overlapping instances to address dataset imbalance. SMOTEENN, on the other hand, merges SMOTE with edited nearest neighbors (ENN) to improve class balance through synthetic sample generation and noise reduction. The comparative analysis reveals that both methods significantly enhance classification performance, SMOTETomek-XGB consistently outperforms SMOTEENN-XGB across key evaluation metrics, including accuracy, F1 score, recall, and receiver operating characteristic - area under the curve (ROC-AUC), underscoring its superior effectiveness in distinguishing phishing sites from legitimate ones. This study offers practical insights into the application of advanced resampling methods for improving machine learning model performance in cybersecurity.
Volume: 15
Issue: 3
Page: 2935-2945
Publish at: 2026-06-01

Brain tumor detection using VGG-16 model

10.11591/ijai.v15.i3.pp2337-2346
Aicha Oussous , Abderrahmane Ez-zahout , Soumia Ziti
Research in medical image analysis, specifically through deep convolutional networks, addresses the challenges of manually analyzing large magnetic resonance imaging (MRI) image volumes for brain tumor detection. The manual analysis is time-consuming, tedious, and prone to inaccuracies due to subtle visual similarities between normal tissue and tumor cells. This research aims to automate tumor detection, increasing accuracy and efficiency in medical treatments. This study aimed to develop a model capable of classifying brain tumors 2D MRI images, and the convolutional neural network (CNN)-based model successfully achieved an accuracy of 99.21% but suffered from noticeable Overfitting. Implementing the independent tests set and early stopping mitigated this issue, making the model more reliable for production deployment and demonstrating its potential in supporting physicians in detecting brain tumors, thereby enhancing treatment efficiency. The use of Python, TensorFlow, and Keras facilitated the development of the proposed solution, focusing on a diverse set of MRI images with varying tumor sizes, locations, shapes, and intensities.
Volume: 15
Issue: 3
Page: 2337-2346
Publish at: 2026-06-01

Pneumothorax detection using a learning focal point architecture

10.11591/ijai.v15.i3.pp2041-2052
Salah-Eddine Mansour , Bouabid Qabliyane , Abdelhak Sakhi , Zakaria Khoudi , Mohamed Baslam
Automatic image segmentation and feature analysis play a crucial role in improving the accuracy and efficiency of disease diagnosis and treatment within modern medical practice. This study propose the use of the learning focal point (LFP) architecture, which is based on the LFP algorithm, to perform effective segmentation of medical images by dividing each image in the dataset into multiple meaningful zones. This zonal segmentation strategy enables the precise extraction of critical regions of interest that are most relevant for pathological analysis. The proposed approach is specifically applied to the detection of common pneumothorax in lung imaging, a condition that requires timely and accurate diagnosis. By concentrating on essential lung zones, the LFP architecture enhances the reliability and robustness of pneumothorax identification. The results demonstrate that this method has the potential to significantly assist clinicians by providing more accurate diagnostic support and facilitating earlier medical intervention, ultimately improving patient outcomes.
Volume: 15
Issue: 3
Page: 2041-2052
Publish at: 2026-06-01

Machine learning approaches for anomaly detection of Jakarta air quality index

10.11591/ijai.v15.i3.pp2543-2553
Muhammad Rizky Nurhambali , Yenni Angraini , Anwar Fitrianto
Anomalies in time series data are observations that deviate markedly from surrounding values or overall patterns. Air quality index (AQI) data, which vary over time, provide a suitable context for anomaly detection. Time series anomaly detection can be done with machine learning approaches like long short-term memory (LSTM) and extreme gradient boosting (XGBoost). These methods have advantages over conventional methods in handling nonlinearity and large data dimensions. This study compares LSTM and XGBoost methods for detecting anomalies in Jakarta's hourly AQI data. The dataset was obtained from the AirNow website and covers the period from January 1, 2018, to December 31, 2023. Anomalies in the observed data were labeled using moving range (MR) (2) and (3) approaches with three and four-sigma thresholds, and feature engineering (FE) was applied to improve model performance. The results indicate that LSTM is more suitable than XGBoost for forecasting and classification tasks in AQI data. LSTM achieved an average mean absolute percentage error (MAPE) of 10.3840%, a root mean square error (RMSE) of 10.5913, and a balanced accuracy (BACC) of 0.9424 under MR (2) labeling with the four-sigma rule. The anomalies detected mostly occurred between 21:00 and 09:00 and during the rainy season.
Volume: 15
Issue: 3
Page: 2543-2553
Publish at: 2026-06-01

Image enhancement combined with EfficientNet-B7 for grading classification of diabetic retinopathy

10.11591/ijai.v15.i3.pp2554-2566
Nina Sevani , Edy Kristianto , Albert Salomo
Diabetic retinopathy (DR) is a complication caused by poorly managed diabetes that affects the eyes. According to the World Health Organization (WHO), 422 million people worldwide have suffered from DR in the past ten years. Manual detection using retinal fundus images is time-consuming and requires experienced ophthalmologists. This study proposes a deep learning method using the pre-trained model EfficientNet-B7 to identify this disease automatically. Five levels of DR will be classified: no-DR, mild-DR, moderate-DR, severe-DR, and proliferative-DR. The model was trained using "APTOS 2019 blindness detection" dataset, and image augmentation was performed. Image segmentation techniques such as contrast limited adaptive histogram equalization (CLAHE) and real enhanced super resolution generative adversarial network (Real-ESRGAN) were applied during preprocessing to improve the model's accuracy significantly. The implementation of CLAHE resulted in the validation accuracy improvement from 76.6% to 83.4% compared to no segmentation, while the combination of Real-ESRGAN and CLAHE increased the accuracy to 93.7%. Future research can explore the combination of CLAHE with other image processing techniques apart from the Real-ESRGAN model.
Volume: 15
Issue: 3
Page: 2554-2566
Publish at: 2026-06-01

Fine-tuning convolutional neural network for artificial intelligence generated image detection enhancement

10.11591/ijai.v15.i3.pp2238-2246
Steven Vincent Hendrawan , Moeljono Widjaja , Alethea Suryadibrata
The relationship between art and technology has changed how people engage with creativity, leading to the industrialization of the field. Various digital media have been utilized in the endeavour of art creation, such as artificial intelligence (AI) generation for images. The utilization of AI-generated art has yielded negative reactions due to its exploitative nature on pre-existing artworks without the creator’s consent, which raises plagiarism concerns. This research utilized convolutional neural network (CNN) to help detect such images to reduce public concerns on the abuse of AI images. The algorithm is proposed to detect such images as it involves spatial convolution within two-dimensional spaces, matching the nature of images. The model was developed from pre-existing architectures, namely EfficientNetB1 and Xception, which was pre-trained on ImageNet classification task with the modification of inclusion or exclusion of dropout in the top layer. After assessing the models, removing top layer dropout from EfficientNetB1 model improved it to reach the F1-score of 97.66% compared to 97.44% in the base model and Xception with a dropout layer yields lower F1-score of 95.56% compared to 97.07% in the base model.
Volume: 15
Issue: 3
Page: 2238-2246
Publish at: 2026-06-01

Artificial intelligence-based risk assessment in agro-industry using supervised neural networks

10.11591/ijai.v15.i3.pp2260-2268
Imam Santoso , Izzum Wafi'uddin , Naila Maulidina Lu'ayya , Annisa'u Choirun , Siti Asmaul Mustaniroh , Dodyk Pranowo , Ainur Rofiq
The coffee supply chain involves high production volumes, complex multi actor interactions, and increasing sustainability requirements, yet remains highly vulnerable to risks dimension. This study aims to develop and evaluate a decision-support framework that improves the accuracy and consistency of sustainability risk classification in the coffee supply chain. The proposed framework integrates failure mode and effect analysis (FMEA) with a supervised artificial neural network (ANN) using backpropagation (BP) to enable data-driven and adaptive risk assessment. Empirical data was collected from 55 respondents, resulting in the identification of 35 supply chain risk factors. These data were used to train and validate an ANN-based classification model implemented in a Python environment, with standard preprocessing and stratified data partitioning to ensure robustness. The ANN classified risks into five categories using supervised learning. The results demonstrate strong predictive performance, achieving overall accuracy of 98.97%, with precision, recall, and F1-scores exceeding 96.8% across all risk classes. Confusion matrix analysis confirms reliable generalization and minimal misclassification. The findings indicate that integrating FMEA with ANN-BP significantly enhances risk classification compared to conventional qualitative approaches. The proposed framework provides a scalable and reliable decision-support tool for dynamic risk scoring, supporting enhancement of sustainable practices in agro-industrial coffee supply chains.
Volume: 15
Issue: 3
Page: 2260-2268
Publish at: 2026-06-01

Deep hybrid models for bitcoin forecasting: EMD, CEEMDAN,and LSTM in comparison

10.11591/ijai.v15.i3.pp2797-2810
Ayoub Aarabi , Maryem Ait Moulay , Issam Bouganssa , Abdelali Lasfar
In this study, an artificial neural network (ANN) was developed to forecast Bitcoin prices using one of the most successful deep learning architectures for time series analysis: long short-term memory (LSTM) networks. This model was enhanced with a signal processing layer that reduces the impact of the instrument’s high volatility on prediction accuracy by applying two signal decomposition techniques: empirical mode decomposition (EMD) and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). This study is motivated by the major fluctuations in Bitcoin prices, which make precise forecasting difficult but crucial for experts and investors. This findings demonstrate that forecasting performance improves when decomposition techniques are used. In particular, compared to the conventional LSTM and EMD-LSTM models, the CEEMDAN-LSTM model achieved the highest accuracy, with a mean absolute error (MAE) of 167.837 and a root mean square error (RMSE) of 255.673, outperforming both EMD-LSTM (MAE =168.785, RMSE =256.042) and the standard LSTM (MAE =169.516, RMSE=256.225). The combination of CEEMDAN and LSTM results in a more reliable model that can accurately capture short-term fluctuations in Bitcoin prices.
Volume: 15
Issue: 3
Page: 2797-2810
Publish at: 2026-06-01

Adaptive multi-scale convolutional network for plant leaf disease detection and classification

10.11591/ijai.v15.i3.pp2956-2969
Tejashwini C. Gadag , D. R. Kumar Raja
Plant disease detection is a critical task in modern agriculture, directly impacting crop yield, food security, and sustainable farming practices. Traditional methods rely on expert visual inspection, which is time consuming, inconsistent, and inaccessible in remote areas. This study introduces advanced deep learning (DL) framework, the adaptive multi-scale convolutional network (AMS-ConvNet) optimized for accurate and efficient plant disease identification. Hierarchical feature extraction network (HFEN) integrates the multi-domain attention framework (MDAF) and adaptive scale fusion module (ASFM) to enhance feature extraction and address challenges such as complex natural backgrounds, non-uniform leaf structures, and varying environmental conditions. The proposed framework employs pre trained knowledge adaptation (PTKA) techniques to improve generalization and overcome data scarcity. Comprehensive evaluations on multiple datasets demonstrate the model's better performance, achieving state-of-the-art metrics in precision, recall, F1-score, and accuracy. Furthermore, this approach ensures scalability and adaptability, making it suitable for real field conditions. The study emphasizes the importance of robust, automated solutions in minimizing crop losses, reducing labor costs, and enhancing agricultural sustainability through precision disease management.
Volume: 15
Issue: 3
Page: 2956-2969
Publish at: 2026-06-01

A mHealth adoption model for diabetes self-management: patient-centered insights from UNRWA clinics

10.11591/ijict.v15i2.pp553-564
Saleem Mohammad Faraj , Haw Yuan Kang , Raja Rina Raja Ikram , Lizawati Salahuddin
This study develops and validates a mobile health (mHealth) adoption model to enhance diabetes self-management among type 2 diabetes mellitus (T2DM) patients in UNRWA primary healthcare clinics across Palestinian refugee camps. This study fills a gap in research on mHealth adoption in low-resource settings by combining the technology acceptance model (TAM), task-technology fit (TTF), and self-efficacy theory (SET). A descriptive, cross-sectional design was employed using a structured, validated questionnaire administered to 503 T2DM patients. Reliability analysis yielded high internal consistency (Cronbach’s α = 0.808–0.966). Structural equation modeling (SEM) using SPSS and AMOS validated the model fit, evidenced by a comparative fit index (CFI) of 0.941 and a root mean square error of approximation (RMSEA) of 0.048. Out of the eleven factors that were examined, Perceived Usefulness had the most positive impact on self-care management (β = 0.67, p < 0.001), while Task Requirement had the least. Notably, Perceived Self-Efficacy showed no significant effect on behavior (p > 0.05). These findings highlight usability, usefulness, and tool functionality as central to promoting mHealth use. The validated model can be modified for other chronic disease settings in comparable healthcare environments and provides practical advice for creating patient-centered mHealth interventions.
Volume: 15
Issue: 2
Page: 553-564
Publish at: 2026-06-01

Forecasting world sugar contract futures using long short-term memory technique with multi-step ahead forecasting strategy

10.11591/ijai.v15.i3.pp2633-2642
Khairil Anwar Notodiputro , Kayla Fakhriyya Jasmine , Indahwati Indahwati , Wandee Wanishsakpong
Time series analysis using stochastic and dynamic models for data forecasting is a key in assisting planning and decision-making processes in various sectors. Long short-term memory (LSTM), with its advantage in understanding patterns and non-linearity in sequential data, is applied in a multi-step ahead forecasting strategy on world sugar futures prices. Fluctuations in sugar prices have a significant impact on the agriculture, trade, and food industry sectors. Forecasting sugar prices becomes a crucial tool for industries, investors, and traders to anticipate changes and make informed decisions. The objectives of this study are to identify the best strategy for forecasting the world sugar contract price and to perform forecasting using the best model. The research results indicate that hyperparameter tuning in LSTM models produces varied combinations and effects. Furthermore, the recursive strategy is suitable for long-term forecasting, while the direct strategy is appropriate for short-term forecasting. Forecasting values for long-term periods remains challenging in achieving high accuracy.
Volume: 15
Issue: 3
Page: 2633-2642
Publish at: 2026-06-01

Utilization of depth-wise and spatially separable convolutional network fusion for classification of white blood cells

10.11591/ijai.v15.i3.pp2595-2605
Firas Muneam Bachay , Ali Abbas Alzaheiree , Hassenien Ali Hussein , Ahmed Nooruldeen Alsafi , Mohammed Hasan Abdulameer
White blood cells (WBCs) are an essential part of the human immune system, playing a significant role in fighting diseases and infections. Their detection and classification from microscopic blood images is a crucial step in diagnosing various diseases. Looking at cells by hand is still key, but it takes a lot of work and mistakes can happen. So, this study tries to improve how to find and sort WBCs using some cool computer tricks. The study tackling issues like cells being on top of each other, looking different, and not having a ton of data. To achieve this, image enhancement techniques were applied using contrast enhancement algorithm, contrast-limited adaptive histogram equalization (CLAHE), and image segmentation techniques using color isolation are employed, which contributes to more accurate separation of overlapping cells, and enables faster and more efficient diagnosis. To efficiently complete the classification process after the segmentation process, a neural network structure consisting of combining three types of convolutional layers (depthwise, spatially, and convolution) was used. To evaluate the proposed technique, experiments were conducted using an open-source blood cell count and detection (BCCD) dataset from the Kaggle platform, and resulted in achieving a classification accuracy of 99.06% and an F1-score of 99.05%. This highlight of the model’s ability to efficiently deal with the challenges associated with WBC classification.
Volume: 15
Issue: 3
Page: 2595-2605
Publish at: 2026-06-01

Adaptive synthetic-based arrhythmia classification using machine learning techniques

10.11591/ijai.v15.i3.pp2398-2409
Md. Rabiul Islam , Tapan Kumar Godder , Ahsan Ul-Ambia , Ferdib Al-Islam , Bulbul Ahamed , Anindya Nag , Ariful Islam
Cardiac arrhythmias, characterized by irregular heart rhythms, pose a significant challenge for timely and accurate diagnosis. This paper presents an advanced framework for arrhythmia classification that addresses the class imbalance issue using Adaptive Synthetic (ADASYN) sampling combined with a comprehensive set of machine learning techniques. We implemented various classifiers, including Logistic Regression, Naive Bayes, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Multi-Layer Perceptron, Gradient Boosting, AdaBoost, Light Gradient Boosting, CatBoost, and Extreme Gradient Boosting. Our experimental results demonstrate that Random Forest, Gradient Boosting, Light Gradient Boosting, and XGBoost achieved a remarkable accuracy of 97%. Other models, such as Decision Tree and Logistic Regression, also performed well, achieving 95% and 94% accuracy, respectively. KNN and Naive Bayes yielded 93% and 81% accuracy, respectively, while AdaBoost underperformed with an accuracy of 24%. Precision scores across the models remained high, except for Naive Bayes and AdaBoost. All models, except AdaBoost, demonstrated excellent recall. Our proposed methodology outperforms previous works, setting a new benchmark for arrhythmia classification. These findings emphasize the effectiveness of integrating ADASYN with machine learning techniques to enhance arrhythmia detection, with significant potential for improving clinical diagnostic processes and patient outcomes.
Volume: 15
Issue: 3
Page: 2398-2409
Publish at: 2026-06-01

Mitigating gender bias in STEM study field classification using GRU and LSTM with augmented dataset technique

10.11591/ijict.v15i2.pp447-455
Devi Fitrianah , Sarah Safitri , Nadzla Andrita Intan Ghayatrie
This study examines gender bias in artificial intelligence (AI), focusing on the classification of high school students into science, technology, engineering, and mathematics (STEM) and non-STEM fields. Using Indonesian student Computer Science Department, BINUS Graduate Program – Master of Computer Science, Bina Nusantara University, Jakarta, 11480 data, conditional variational autoencoder (CVAE) and multilabel synthetic minority over-sampling technique (MLSMOTE) were employed for data augmentation to mitigate bias before training gated recurrent unit (GRU) and long short-term memory (LSTM) models for prediction. The combination of MLSMOTE and GRU demonstrated superior performance, achieving accuracies of 93% for female students and 94% for males. These results indicate that MLSMOTE and GRU effectively predict fields of study while addressing gender bias. The findings contribute to advancing fairness in AI systems for education and beyond, ensuring equitable opportunities across diverse applications.
Volume: 15
Issue: 2
Page: 447-455
Publish at: 2026-06-01
1 2 3
...
2091 2092
Show 62 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