International Journal of artificial intelligence research (IJAIR)
Not a member yet
271 research outputs found
Sort by
Development of a Web-Based Aviation English Proficiency Test: Integrating Adaptive Algorithms and Dynamic Assessment for Enhanced Evaluation in Aviation Education
Full text linkAviation English proficiency is pivotal for aviation school students to ensure secure communication in global airspace per ICAO guidelines. Conventional methods are rigid, leading to inaccurate and time-consuming evaluations that hinder training efficacy. This research develops a web-based adaptive Aviation English proficiency test integrating adaptive algorithms like Item Response Theory and dynamic assessment to enhance aviation education outcomes. Using a mixed-methods framework with the ADDIE model and quantitative experimental approach, an explanatory sequential design with non-equivalent control group was employed, involving needs assessment, prototype development, validation, and implementation. The sample included 141 aviation school students. Data from pre/post-tests were analyzed via SPSS. The findings showed that i) the web-based test is valid and feasible as an assessment tool with a validation score of 89.5%; ii) student proficiency levels are significantly improved before and after using the adaptive system (paired t-test: mean rise from 72.6 to 91.4, t=-14.28, p=0.000 <0.05); iii) dynamic assessment positively impacts learning outcomes following implementation (32% uplift, ?=0.61, p<0.01); and iv) there is a significant difference between experimental and control groups in evaluation efficiency (independent t-test: 25% higher for experimental, t=10.52, p=0.000 <0.05). These affirm the test's efficacy, recommending broader adoption for refined aviation training
Salt Quality Classification Using Backpropagation Neural Network and K-Nearest
Full text linkSalt quality plays a vital role in determining its usability across various sectors, including food, pharmaceuticals, and industrial applications. Traditional methods of classifying salt quality, which rely heavily on manual inspection and laboratory testing, are often time-consuming, costly, and prone to human error. In response to these limitations, this study explores the implementation of machine learning techniques—specifically, Backpropagation Neural Network (BPNN) and K-Nearest Neighbor (K-NN)—to classify salt quality based on its physical and chemical properties. The features used in this research include NaCl concentration, moisture content, magnesium levels, sulfat, insoluble, calcium, NaCL(wb) and NaCL(db) which are commonly used indicators of salt purity and grade. The BPNN model is designed to handle complex and non-linear relationships within the dataset by adjusting weights through iterative backpropagation during training. Meanwhile, the K-NN algorithm serves as a simpler, instance-based learning method that classifies samples based on the majority class of their nearest neighbors in the feature space. Comparative experiments were conducted to evaluate the classification and computational efficiency of both models. Results indicate that both methods are effective in classifying salt into predefined quality categories. However, BPNN consistently outperforms K-NN in terms of time efficiency and generalization, particularly when handling noisy or overlapping data. The findings underscore the potential of integrating artificial intelligence into quality control systems in the salt industry, offering a faster, more objective, and scalable solution for ensuring product standards
Explainable AI-Based Real-Time Hybrid System for Blockchain Anomaly Detection: A Multi-Cryptocurrency Perspective
Full text linkThis study achieves a 5% improvement in AUC-ROC and a 2.5% increase in recall compared to state-of-the-art anomaly detection methods in blockchain networks. Blockchain technologies have rapidly evolved, offering transparency and security across decentralized systems. However, detecting anomalies and fraudulent activities remains a significant challenge. This research proposes a unified hybrid framework integrating Graph Neural Networks (GNNs), Transformers, and XGBoost within a federated learning environment for real-time anomaly detection in multi-cryptocurrency blockchain networks. Unlike previous works, this model employs explainable AI (XAI) methods (SHAP and LIME) to enhance interpretability and trust. The framework utilizes PSO-based hyperparameter optimization, reducing convergence time by 20%. Experimental evaluations on benchmark datasets (Elliptic, Bitcoin-OTC, and Ethereum) demonstrate superior performance in precision, recall, and FPR compared to CARE-GNN and GeniePath. The results confirm the proposed model’s scalability, transparency, and real-time efficiency, making it suitable for deployment in high-frequency blockchain monitoring systems.
Comparative Study of Fuzzy Inference System and Adaptive Neuro-Fuzzy Inference System in Public Sentiment Analysis of Kabinet Merah Putih
No full textPurpose: This study aims to compare two fuzzy logic-based approaches, namely the Fuzzy Inference System (FIS) and the Adaptive Neuro-Fuzzy Inference System (ANFIS), in analyzing public sentiment toward the Kabinet Merah Putih.Methods: A dataset of 1,197 tweets was collected from Twitter (X) between October 2024 and April 2025 using specific keywords. After preprocessing and polarity measurement with TextBlob, the sentiment values were mapped into seven categories: strongly negative, negative, weakly negative, neutral, weakly positive, positive, and strongly positive. The classification was performed using both FIS and ANFIS. Evaluation metrics included accuracy, precision, recall, F1-score, and error rate (MSE and RMSE).Result: Experimental results show that FIS achieved an overall accuracy of 79.2%, performing well on majority classes but failing to identify several minority classes. In contrast, ANFIS obtained an accuracy of 92.5% with very low error (MSE = 0.0341, RMSE = 0.1848), demonstrating strong capability in classifying majority and several minority categories. Overall, ANFIS outperformed FIS, proving more effective in capturing sentiment patterns and aligning with the actual distribution of public opinion..Novelty: This study offers novelty by explicitly comparing the performance of FIS and ANFIS in multi-level sentiment analysis of Indonesian social media data, an approach that has not been explored in prior research
IoT-Based Monitoring for Optimizing Yield of Gogo Rice (Oryza sativa, L.)
Full text linkAdvancements in Internet of Things (IoT) technology have introduced new opportunities in precision agriculture, particularly for enhancing the productivity of upland rice (Oryza sativa, L.) cultivated on marginal lands. This study aims to integrate an IoT-based monitoring system with the application of biochar and Trichoderma harzianum to optimize soil parameters and water resource efficiency. The monitoring system utilizes Trico Master and Slave devices to measure real-time environmental parameters, including soil pH, soil moisture, soil temperature, and air temperature. The results reveal that the application of biochar at a dosage of 1 kg/m² increased soil pH from an average of 7.0 to 8.7, creating a conducive environment for the activity of Trichoderma harzianum. This microorganism demonstrated its ability to improve soil quality by decomposing organic matter and enhancing nutrient absorption by plants. Additionally, the IoT-based automated irrigation system maintained soil moisture levels above 45% while reducing water usage by up to 30% compared to manual irrigation methods. In conclusion, the integration of IoT technology with biochar and Trichoderma harzianum significantly improved upland rice yield, resource efficiency, and the sustainability of agricultural systems. This study presents an innovative and sustainable approach to supporting future food security, particularly in resource-limited environment
Legal Policies for the Protection of Children's Rights After Divorce: A Comparative Study of Malaysia, Singapore, and Indonesia
Full text linkDivorce has a significant impact on children, particularly concerning the fulfillment of their custody, financial support, and psychosocial well-being. This study analyzes legal policies on the protection of children's rights after divorce in indonesia, malaysia, and singapore through a comparative law approach. Using normative and empirical juridical methods, this research examines existing legal regulations and the effectiveness of their implementation based on court case data, divorce statistics, and interviewxs with legal experts. The findings indicate that while all three countries have regulations ensuring children's rights post-divorce, gaps remain in their implementation. In indonesia, weak enforcement mechanisms for child support lead to many children losing their financial entitlements. Malaysia’s dual legal system (sharia and civil) sometimes delays the execution of child support rights. Meanwhile, singapore has developed an integrated system with a therapeutic justice approach, including mandatory mediation and more effective enforcement mechanisms. This study recommends establishing a child support enforcement unit in indonesia, harmonizing legal procedures in malaysia, and enhancing psychosocial support programs in singapore. Legal reforms incorporating best practices from these three countries are expected to improve the protection of children's rights after divorce
Harnessing Generative AI for ESP: A Cross-Disciplinary Vocational Education Framework with Predictive Modeling Evidence from Indonesia
Full text linkThis study examines how Generative AI tools, specifically ChatGPT and Gemini, can enhance English for Specific Purposes (ESP) learning and education. Drawing on the UTAUT2 model of technology acceptance and recent discussions on AI-mediated learning, we examine the roles of baseline ability, perceived usefulness, and satisfaction as mediating factors in ESP classrooms. Data were collected from 50 vocational students across five departments using pre- and post-tests, AI usage logs, and Likert-scale surveys. Statistical analyses included descriptive statistics, paired t-tests, ANOVA with Tukey adjustment, correlation, reliability tests, and predictive modeling (OLS and LASSO) in SAS Studio. Results show a mean learning gain of 24.42 points, with Nursing and IT students benefiting most. AI usage hours strongly correlate with post-test scores but not directly with learning gain, suggesting that perceived usefulness and satisfaction (both rated 4.4/5 with ? = 1.00) mediate the outcomes. Baseline competence remains the strongest predictor, highlighting persistent disparities in skill distribution across vocational fields. These findings suggest that the effective integration of Generative AI in ESP requires scaffolding and domain-specific alignment, rather than simple exposure. The study offers a novel framework for AI-supported ESP instruction, providing practical guidance for educators and policymakers in Indonesia and similar contexts
Transfer Learning-Based Classification of Herbal Plants for Biodiversity Conservation
No full textIndonesia's status as a megabiodiversity country is threatened by land degradation, endangering its rich herbal plant species. The identification of these plants, crucial for ecological balance and traditional medicine, remains reliant on slow and subjective manual methods. This research addresses this problem by designing an automated, accurate, and accessible classification system for Indonesian herbal plants. The primary objective was to develop and evaluate a deep learning model based on transfer learning with the EfficientNet-B0 architecture, optimized for deployment on mobile devices. The methodology involved curating a dataset of 4,500 images across 15 species from Lombok, validated by botanists. The model was trained using a 70:15:15 data split, with extensive data augmentation (rotation, flipping, zooming) to improve generalization. Experimental results demonstrated that the proposed EfficientNet-B0 model achieved a high classification accuracy of 92.4% and an F1-score of 91.7%, outperforming baseline models like MobileNetV2 and ResNet50. The model was successfully converted to TensorFlow Lite and integrated into an Android application, which facilitates real-time identification with an average inference time of less than one second. The study concludes that the implemented system provides a robust tool for biodiversity conservation and community health initiatives. The main contribution lies in the creation of a specialized dataset, the optimization of EfficientNet-B0 for local herbs, and the development of a functional mobile application for real-world use
Priority Strategy for Goat and Sheep Farming Development using IPA
Full text linkDesigning a business model for sustainable goat and sheep farming is one of the keys to increasing productivity and competitiveness of the livestock sector in Indonesia. This article examines the formulation of a business model based on upstream-downstream integration using the Importance Performance Analysis (IPA) tool to identify priorities that need more attention in the development of goat and sheep farming. The IPA matrix is used to evaluate important components based on the Grand Design for National Goat and Sheep Development 2045 and the Guidelines for the Implementation of the Goat and Sheep Corporation Development Program of the Directorate General of PKH, Ministry of Agriculture. The results of the analysis show that aspects of providing quality feed, strengthening cultivation management, and developing market communication and education are priority areas that need to be optimized. In addition, market demand analysis and logistics systems also need to be improved to support efficient and timely product distribution. Formulating a business model that integrates the entire agribusiness chain can increase efficiency, strengthen competitiveness, and support national food security
LSTM Model Using Adam’s Optimizer for Indonesian – Bugis Bidirectional Translation System
Full text linkThe purpose of this research is to develop a machine translation of Bugis to Indonesian and vice versa in order to preserve the Bugis language. This research utilizes a recent dataset consisting of 30,000 Bugis-Indonesian sentence pairs from the online Bible. This research conducts scraping to compile the corpus which is then followed by manual and automatic pre-processing. The method chosen is Neural Machine Translation (NMT) while for training and testing models Long Short-Term Memory (LSTM) is used. The performance of the model is evaluated by Bilingual Evaluation Understudy (BLEU) score to measure the translation accuracy at various epochs. In addition, this study also compared the use of Adam's optimizer with non-optimizer. The results showed that the use of Adam's optimizer significantly improved the performance of the model where at epoch 2000 the model achieved the highest BLEU score of 0.996261 indicating highly accurate translation quality. In contrast, the model without the optimizer showed lower performance. Other results also found that the translation from Bugis to Indonesian was more accurate than from Indonesian to Bugis. This is due to the more balanced word count difference in the Bugis to Indonesian translation, which makes it easier for the model to match words. In conclusion, the use of NMT with Adam optimizer effectively improves the accuracy of two-way translation from Bugis-Indonesian