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

30,411 Article Results

MobileChiliNet: convolutional neural network for chili leaves classification

10.11591/ijai.v14.i5.pp3757-3770
Sayuti Rahman , Marischa Elveny , Marwan Ramli , Dionikxon Manurung
Chili pepper (Capsicum annuum) is an important crop in many countries, including Indonesia, which plays an important role in local economy and food production. To meet the high demand, effective agricultural management, especially the diagnosis and treatment of plant diseases, is essential. This study aims to improve the accuracy of chili leaf disease classification while reducing the computational cost so that it can be applied to low-cost smart farming systems. Through the development of the MobileChiliNet architecture, which is the result of pruning and fine-tuning of MobileNetV2, this model achieves the best accuracy, better than other CNNs such as ResNet50 and VGG16. Testing with various optimizers and learning rate schedulers shows that AdamW with PolynomialDecay provides the best performance by increasing the validation accuracy to 96.48%. This approach successfully reduces the computational complexity while maintaining high accuracy, so that it can be implemented in smart farming systems at a lower cost.
Volume: 14
Issue: 5
Page: 3757-3770
Publish at: 2025-10-01

Object detection for indoor mobile robot: deep learning approaches review

10.11591/ijai.v14.i5.pp3520-3527
Hind Messbah , Mohamed Emharraf , Mohamed Saber
Efficient object detection is crucial for enabling autonomous indoor robot navigation. This paper reviews current methodologies and challenges in the field, with a focus on deep learning-based techniques. Methods like you only look once (YOLO), region-based convolutional neural networks (R-CNN), and Faster R-CNN are explored for their suitability in real-time detection in dynamic indoor environments. Deep learning models are emphasized for their ability to improve detection accuracy and adaptability to varying conditions. Key performance metrics such as accuracy, speed, and scalability across different object types and environmental scenarios are discussed. Additionally, the integration of object detection with navigation systems is examined, highlighting the importance of accurate perception for safe and effective robot movement. This study provides insights into future research directions aimed at advancing the capabilities of indoor robot navigation through enhanced deep learning-based object detection techniques.
Volume: 14
Issue: 5
Page: 3520-3527
Publish at: 2025-10-01

Two-steps feature selection for detection variant distributed denial of services attack in cloud environment

10.11591/ijai.v14.i5.pp3945-3957
Kurniabudi Kurniabudi , Eko Arip Winanto , Sharipuddin Sharipuddin
The prevalence of cloud computing among organizations poses a significant problem in ensuring security. Specifically, distributed denial of services (DDoS) attacks targeting cloud computing networks can lead to financial losses for consumers of cloud computing services. This assault has the potential to render cloud services inaccessible. The detection system serves as a remedy to prevent more substantial losses. This research aims to enhance the efficacy of the system detection model by integrating feature selection with three machine learning algorithms: decision tree (DT), random forest (RF), and naïve Bayes (NB). Therefore, our study suggests combining two phases of feature selection into the DDoS attack detection procedure. The first phase uses the information gain (IG) feature selection technique approach, and the second phase uses the principal component analysis (PCA) feature extraction approach. The technique is referred to as two-step feature selection. The test findings indicate that the implementation of two-step feature selection can enhance the performance of the DT and RF detection models by around 9%.
Volume: 14
Issue: 5
Page: 3945-3957
Publish at: 2025-10-01

Ensemble reverse knowledge distillation: training robust model using weak models

10.11591/ijai.v14.i5.pp4162-4170
Christopher Gavra Reswara , Tjeng Wawan Cenggoro
To ensure that artificial intelligence (AI) can be aligned with humans, AI models need to be developed and supervised by humans. Unfortunately, it is possible for an AI to exceed human capabilities, which is commonly referred to as superalignment models. Thus, it raised the question of whether humans can still supervise a superalignment model, which is encapsulated in a concept called weak-to-strong generalization. To address this issue, we introduce ensemble reverse knowledge distillation (ERKD), which leverages two weaker models to supervise a more robust model. This technique is a potential solution for humans to manage a super-alignment of models. ERKD enables a more robust model to achieve optimal performance with the assistance of two weaker models. We tried to train a more robust EfficientNet model with weaker convolutional neural network (CNN) models in a supervised fashion. With this method, the EfficientNet model performed better than the model trained with the standard transfer learning (STL) method. It also performed better than a model that was supervised by a single weaker model. Finally, ERKD-trained EfficientNet models can perform better than EfficientNet models that are one or even two levels stronger.
Volume: 14
Issue: 5
Page: 4162-4170
Publish at: 2025-10-01

Optimized ensemble framework for predicting hydroponic stock and sales using machine learning

10.11591/ijai.v14.i5.pp3879-3886
Viktor Handrianus Pranatawijaya , Ressa Priskila , Putu Bagus Adidyana Anugrah Putra , Nova Noor Kamala Sari , Efrans Christian , Septian Geges , Novera Kristianti
The increasing global demand for food necessitates the adoption of sustainable agricultural practices. Hydroponic farming, while efficient in resource utilization, faces challenges in accurately predicting stock levels and sales due to dynamic, ever-changing factors. This research presents an optimized ensemble framework for forecasting hydroponic stock levels and sales by integrating linear regression (LR), random forest (RF), and XGBoost, further enhanced through an evolutionary algorithm (EA). The proposed framework is evaluated using root mean square error (RMSE) and mean absolute error (MAE), demonstrating significant accuracy improvements over individual models. The ensemble model achieves an RMSE reduction of 43.82% for stock prediction and 55.3% for sales forecasting compared to the best-performing individual model. Additionally, local interpretable model-agnostic explanations (LIME) are employed to offer stakeholders clear insights into decision-making processes, such as identifying "number of harvested crops" and "sales data" as key drivers of prediction outcomes. This framework supports sustainable development goals (SDGs) 9.3, 12.3, and 12.C by promoting resource efficiency, reducing food waste, and improving small-scale farmer market access. Future research will explore real-time data integration for dynamic adaptation and further model enhancements.
Volume: 14
Issue: 5
Page: 3879-3886
Publish at: 2025-10-01

Enhanced classification of aromatic herbs using EfficientNet and transfer learning

10.11591/ijai.v14.i5.pp4123-4136
Samira Nascimento Antunes , Madalena De Oliveira Barbosa Divino , Luana Dos Santos Cordeiro , Fernanda Pereira Leite Aguiar , Marcelo Tsuguio Okano
Herbs have long been used for culinary and medicinal purposes, as well as in religious rituals, due to their essential oils and aromatic properties. However, distinguishing between aromatic and medicinal herbs based on visual characteristics alone can be challenging. With recent advances in computer vision, plant identification from images has seen significant growth, offering promising applications in several domains. This article aims to evaluate the classification of aromatic herbs using the EfficientNet convolutional neural network (CNN) technique with transfer learning. The methodology used is experimental research, systematically manipulating variables to observe their effects on the object of study. The researcher plays an active role in this process, rather than being a passive observer. Based on the results and the literature review, it is evident that the objective of this research was achieved, as despite the opportunities for improvement in training to achieve accuracy above 0.8, it was possible to evaluate the classification of aromatic herbs using EfficientNet CNN through the transfer learning technique.
Volume: 14
Issue: 5
Page: 4123-4136
Publish at: 2025-10-01

Residual edge dense enhanced module network: a deep learning approach with multi-class SVM for lung tumor stage classification

10.11591/ijai.v14.i5.pp4032-4042
Prabakaran Jayaraman , Pandiaraj Selvaraj , Ashwini Elango
Lung cancer segmentation with positron emission tomography (PET) and computed tomography (CT) images plays a critical role to accurately detect lung cancer. Nevertheless, lung tumor segmentation in PET/CT images were extremely difficult due to the movement caused by respiration. Despite this fact, the lung tumor images shown large number of variations mostly in PET images and CT images. As PET-CT images are acquired concurrently the shape and size of lung tumor varies according to modality. To address these issues, we developed a residual edge dense enhanced module network (REDEM-NET) framework for lung tumor stage classification. The proposed REDEM-NET can process PET and CT images as inputs. In addition, the dense residual convolutional network (DRCN) collects both inputs and extracts high-dimensional features concurrently. The extracted features from both imaging modalities were fed into UNet+++ to obtain multi-level decoded features. The extracted decoded features are concurrently supplied to the pixel level learning module (PELM) and edge level learning module (E2LM) which resulting in two outputs for subsequent learning. The outputs were merged to provide a very precise lung tumor segmentation. Furthermore, segmented tumor was fed to multi-class support vector machine (MC-SVM) for lung tumor stage classification. Moreover, it was able to identify three stages and its substages namely primary tumor, region lymph node and distant metastasis.
Volume: 14
Issue: 5
Page: 4032-4042
Publish at: 2025-10-01

Comparison of HSV-color and ANN-HSV-color segmentation for detecting soybean adulteration

10.11591/ijai.v14.i5.pp3734-3743
Farid Rahmat Abadi , Rudiati Evi Masithoh , Lilik Sutiarso , Sri Rahayoe
Soybeans are an important food crop, but their quality is often compromised by contamination with other materials, a process known as adulteration. Conventional methods for detecting adulteration are slow; therefore, there is a need for rapid and non-invasive alternatives. This study aimed to assess the capability of hue-saturation-value (HSV) color segmentation and its combination with artificial neural networks (ANN) to identify adulteration in soybean samples. This research employed image processing and machine learning to segment soybeans mixed with adulterants at concentrations of 5%, 10%, 15%, 20%, and 25%. The HSV method successfully distinguished soybeans and other materials, but some challenges were observed in shadow regions and areas with similar colors. The HSV-ANN model with six hidden layers performed well with a calibration accuracy of R² value of 0.97 and root-mean-square error (RMSE) of 2.16%, which provided more detailed segmentation, although it still had some problems in shadow regions and undetected corn embryo parts. The validation results indicated that the HSV model had an R² value of 0.98 and RMSE of 4.48%, while the HSV-ANN model had an R² value of 0.96 and RMSE of 1.3%. Both models were capable of predicting the levels of adulteration, and the HSV-ANN model proved to be more accurate. It is concluded that both methods are efficient; however, there is a need for more work on modeling and sampling to increase the segmentation precision and decrease the biases, especially in the shadow and overlapped color.
Volume: 14
Issue: 5
Page: 3734-3743
Publish at: 2025-10-01

Design and analysis of reinforcement learning models for automated penetration testing

10.11591/ijai.v14.i5.pp4061-4073
Suresh Jaganathan , Mrithula Kesavan Latha , Krithika Dharanikota
Our paper proposes a framework to automate penetration testing by utilizing reinforcement learning (RL) capabilities. The framework aims to identify and prioritize vulnerable paths within a network by dynamically learning and adapting strategies for vulnerability assessment by acquiring the network data obtained from a comprehensive network scanner. The study evaluates three RL algorithms: deep Q-network (DQN), deep deterministic policy gradient (DDPG), and asynchronous episodic deep deterministic policy gradient (AE-DDPG) in order to compare their effectiveness for this task. DQN uses a learned model of the environment to make decisions and is hence called model-based RL, while DDPG and AE-DDPG learn directly from interactions with the network environment and are called model-free RL. By dynamically adapting its strategies, the framework can identify and focus on the most critical vulnerabilities within the network infrastructure. Our work is to check how well the RL technique picked security vulnerabilities. The identified vulnerable paths are tested using Metasploit, which also confirmed the accuracy of the RL approach's results. The tabulated findings show that RL promises to automate penetration testing tasks.
Volume: 14
Issue: 5
Page: 4061-4073
Publish at: 2025-10-01

The effectiveness of ChatGPT in extracting architectural patterns and tactics

10.11591/ijai.v14.i5.pp4363-4370
Hind Milhem , Naderah Al-Jawabrah , Raghad Abu Wadi
This work investigates the potential of ChatGPT, a cutting-edge large language model (LLM), for software design analysis specifically in detecting architectural patterns and tactics. The evaluation involves comparing ChatGPT’s performance with that of Archie, a traditional Eclipse plugin designed for architectural analysis. The study uses the source code of five open-source software systems as the testing ground. Results reveal that ChatGPT achieves noteworthy performance in both pattern and tactic detection tasks. Specifically, for pattern detection, ChatGPT demonstrates an accuracy of up to 47.06%, while for tactic detection, it achieves a precision of 28.25%. While ChatGPT’s current capabilities are not yet a replacement for specialized tools like Archie, it offers significant potential as a complementary tool in architectural analysis workflows. By bridging the gap between natural language understanding and software engineering, ChatGPT could pave the way for more intelligent and automated solutions in the field. However, a key limitation is its difficulties in handling foundational or traditional tactics, resulting in a lower detection rate in certain areas. This research contributes valuable insights into the application of LLMs in software engineering, highlighting both the strengths and the limitations of ChatGPT in addressing complex architectural tasks.
Volume: 14
Issue: 5
Page: 4363-4370
Publish at: 2025-10-01

Detection of chronic kidney disease based on ensemble approach with optimal feature selection using machine learning

10.11591/ijai.v14.i5.pp4017-4031
Deepika Amol Ajalkar , Jyoti Yogesh Deshmukh , Mayura Vishal Shelke , Shalini Vaibhav Wankhade , Shwetal Kishor Patil
Chronic kidney disease (CKD) poses a significant health risk globally, necessitating early and accurate detection to ensure timely intervention and effective treatment. This study presents an advanced ensemble machine learning (ML) approach combined with optimal feature selection to enhance the detection of CKD. Using five baseline ML classifiers like gradient boosting (GB), random forest (RF), K-nearest neighbors (KNN), support vector machine (SVM), and decision tree (DT), and utilizing grid search for hyperparameter tuning, the proposed ensemble model capitalizes on the strengths of each algorithm. Our approach was tested on a public benchmark CKD dataset from Kaggle. The experimental results demonstrate that the ensemble model consistently outperforms individual classifiers and existing methods, achieving 97.5% accuracy, precision, recall, and an F1-score of 97.4%. This superior performance underscores the ensemble model's potential as a reliable early CKD detection tool. Integrating ML into CKD diagnostics enhances accuracy. It facilitates the development of automated, scalable diagnostic tools, aiding healthcare professionals in making informed decisions and ultimately improving patient outcomes.
Volume: 14
Issue: 5
Page: 4017-4031
Publish at: 2025-10-01

Optimizing nitik batik classification through comparative analysis of image augmentation

10.11591/ijai.v14.i5.pp3970-3981
Suprapto Suprapto , Meilany Nonsi Tentua , Ahmad Rizki Maulana
Nitik batik is one of the most intricate and culturally significant motifs in Yogyakarta's batik tradition, characterized by its complex, geometric dot-based patterns. The unique challenges of automatically classifying nitik batik motifs stem from the high variability within the class and the limited availability of training data. This study investigates how different image data augmentation techniques can enhance the performance of a random forest classifier for nitik batik motifs. Techniques such as geometric transformations (flip, rotate, and scaling), intensity transformations (cut-out, grid mask, and random erasing), non-instance level augmentation (pairing samples), and unconditional image generation (deep convolutional generative adversarial network (DCGAN)) were used to expand the dataset and improve the model's ability to generalize. The results show that specific techniques, notably flip, cut-out, and DCGAN, significantly improved classification accuracy, with flip achieving the highest accuracy improvement of 20.20%, followed by cut-out at 19.27% and DCGAN at 16.25%. Moreover, DCGAN demonstrated the lowest standard deviation (0.78%), indicating high stability and robustness in classification performance across multiple validation folds. These findings suggest that augmentation techniques effectively improve classification accuracy and enhance the model's ability to generalize from limited and complex datasets.
Volume: 14
Issue: 5
Page: 3970-3981
Publish at: 2025-10-01

Learning assistance module based on a small language model

10.11591/ijai.v14.i5.pp4202-4210
Marco Antonio Jinete , Robinson Jiménez-Moreno , Anny Astrid Espitia-Cubillos
This paper presents the development of a low-cost learning assistant embedded in an NVIDIA Jetson Xavier board that uses speech and gesture recognition, together with a long language model for offline work. Using the large language model (LLM) Phi-3 Mini (3.8B) model and the Whisper (model base) model for automatic speech recognition, a learning assistant is obtained under a compact and efficient design based on extensive language model architectures that give a general answer set of a topic. Average processing times of 0.108 seconds per character, a speech transcription efficiency of 94.75%, an average accuracy of 9.5/10 and 8.5/10 in the consistency of the responses generated by the learning assistant, a full recognition of the hand raising gesture when done for at least 2 seconds, even without fully extending the fingers, were obtained. The prototype is based on the design of a graphical interface capable of responding to voice commands and generating dynamic interactions in response to the user's gesture detection, representing a significant advance towards the creation of comprehensive and accessible human-machine interface solutions.
Volume: 14
Issue: 5
Page: 4202-4210
Publish at: 2025-10-01

Intent detection in AI chatbots: a comprehensive review of techniques and the role of external knowledge

10.11591/ijai.v14.i5.pp4250-4259
Jemimah K. , Rajkumar Kannan , Frederic Andres
Artificial intelligence (AI) chatbots have become essential across various industries, including customer service, healthcare, education, and entertainment, enabling seamless, and intelligent user interactions. A key component of chatbot functionality is intent detection, which determines the underlying purpose of user queries to provide relevant responses. Traditional intent detection methods, such as rule-based and statistical approaches, often struggle with adaptability, especially in complex, dynamic conversations. This review examines the evolution of intent detection techniques, from early methods to modern deep learning and knowledge-enriched models. It introduces the domain type-conversation turns-adaptivity-external knowledge (DCAD) classification, highlighting its significance in improving chatbot accuracy and contextual awareness. The paper categorizes existing intent detection models, analyzes their applications across various sectors, and discusses key challenges, including data integration, language ambiguity, and ethical concerns. By exploring emerging trends and future directions, this review underscores the critical role of external knowledge in enhancing chatbot performance and user experience.
Volume: 14
Issue: 5
Page: 4250-4259
Publish at: 2025-10-01

Unveiling critical features for failure prediction in green internet of things applications

10.11591/ijai.v14.i5.pp4308-4318
Ouiam Khattach , Omar Moussaoui , Mohammed Hassine
The rapid growth of the green internet of things (GIoT) in recent years signifies a transformative shift in internet of things (IoT) solution development. This evolution is driven by technological advancements, heightened environmental awareness, and a global imperative to combat climate change. Ensuring the reliability of GIoT applications is crucial for their success. This study identifies critical features for predicting IoT device failures, enabling early detection and intervention. Using datasets from industry, energy, and agriculture sectors, we employ a feature selection strategy to analyze extensive data from diverse GIoT deployments. Our analysis identifies significant features and integrates key insights from existing literature. Our findings support enhanced predictive maintenance strategies, reduced downtime, and improved overall performance of sustainable IoT solutions.
Volume: 14
Issue: 5
Page: 4308-4318
Publish at: 2025-10-01
1 2 3
...
2027 2028
Show 144 of 2028

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