Publication Search

This study discusses a student concentration detection system using Convolutional Neural Network (CNN) with the MobileNetV2 architecture. The dataset was adapted from Classroom Student Behaviors and mapped into four concentration categories: highly focused, focused, less focused, and unfocused. The system was tested with a 720p webcam and produced real-time detection data. The evaluation results show an overall accuracy of 75.85%, with the highest precision achieved in the focused class (0.9859) and the highest recall in the highly focused (0.9739) and unfocused (0.9811) classes. The confusion matrix indicates that the focused class was detected most consistently, while highly focused and unfocused classes were often misclassified as focused, resulting in lower precision. In real-time testing, the system operated at an average of 7 FPS and worked optimally when students faced the camera directly with sufficient lighting, but its performance decreased significantly at face angles greater than 45°. User evaluation shows that 75% of students rated the detection results as accurate/very accurate with an average satisfaction score of 3.6 out of 5, and 75% felt assisted in recognizing their concentration level. From the teachers’ perspective, most stated that the results were consistent with classroom observations, and all expressed willingness to reuse the system.

Predictive maintenance (PdM) plays a crucial role in modern industrial systems by minimizing downtime, reducing maintenance costs, and optimizing asset performance. However, many predictive models operate as “black box” systems, limiting transparency and making it difficult for operators to interpret their outputs. This study aims to integrate Explainable Artificial Intelligence (XAI) techniques with Remaining Useful Life (RUL) prediction models to improve both accuracy and interpretability. Various machine learning and deep learning approaches, including Support Vector Machines (SVM), Random Forest (RF), XGBoost, Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN), are employed to predict RUL using real-time sensor data from rotating machinery. XAI methods such as SHAP, LIME, and attention mechanisms are applied to provide human-understandable explanations of model predictions. The models are evaluated based on accuracy, Root Mean Square Error (RMSE), and interpretability scores. The results show that XAI-enhanced models outperform traditional approaches in predictive performance while offering greater transparency. These explanations help maintenance engineers better understand the factors influencing predictions, thereby improving decision-making and trust in the system. Nevertheless, the integration of XAI introduces additional computational complexity, which may pose challenges for large-scale industrial implementation. Overall, this study highlights the potential of combining XAI with RUL prediction to develop more reliable, transparent, and effective predictive maintenance solutions.

This research benchmarks multiple machine learning (ML) algorithms for large-scale loan default prediction using a real-world dataset of 255,000 borrower records, where default cases represent only ~9–12% of total observations. The study addresses the persistent gap in comparative analyses of ML models that balance predictive accuracy, interpretability, and computational efficiency for credit risk assessment. Six algorithmic families were evaluated Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, Artificial Neural Networks (ANN), and Stacked Ensemble—using standardized preprocessing, hybrid imbalance handling (SMOTE, class weighting, under-sampling), and comprehensive evaluation metrics (AUC, F1, Recall, Precision, PR-AUC, and Brier Score). Empirical results show Logistic Regression achieved the highest AUC of 0.732, outperforming nonlinear models under the baseline configuration, while LightGBM attained perfect recall (1.0) but low precision (0.116), indicating over-prediction of defaults. Gradient boosting models demonstrated robust calibration (Brier ≈ 0.114–0.116) and the best computational efficiency, with LightGBM showing the fastest training and lowest memory use. CatBoost exhibited strong recall but the slowest computation, and ANN underperformed on tabular data (AUC ≈ 0.56). The Stacked Ensemble delivered balanced results with AUC = 0.664 and improved overall stability. These findings confirm that boosting-based models, particularly LightGBM and CatBoost, offer superior scalability and calibration, whereas Logistic Regression remains a valuable interpretable baseline. The study concludes that effective default prediction requires integrating rebalancing, calibration, and threshold optimization to enhance recall and operational deployment reliability in large-scale credit ecosystems.

Rainfall is a crucial factor in the stability of the Earth's ecosystem and has a significant impact on agriculture, forestry, energy, and water management. However, increasingly unstable climate change makes rainfall patterns difficult to predict accurately using traditional methods. The city of Medan, the capital of North Sumatra Province, has a tropical rainforest climate with an average annual rainfall of approximately ±2200 mm and an average temperature of 27°C. Significant weather fluctuations in this area can trigger flooding when rainfall increases and cause water shortages when rainfall decreases (BMKG, 2021). Therefore, a prediction approach that can manage non-linear and dynamic data is needed. Artificial Neural Networks (ANN) are one of the reliable machine learning methods for detecting data patterns. By using the backpropagation algorithm, the model can gradually reduce prediction errors, making it widely used in weather forecasting applications. In this regard, this study uses ANN with the backpropagation method to forecast monthly rainfall in Medan City by utilizing data from 2022–2024 as training and testing data.

This study examines the development of a Crude Palm Oil (CPO) price forecasting model using an artificial neural network algorithm, specifically the backpropagation algorithm. As one of Indonesia’s main export commodities, CPO has a significant economic impact and influences the income of oil palm farmers. The CPO price data used in this study were obtained from CIF Rotterdam, covering the period from January 2019 to December 2023. The research methodology consists of several stages, including data collection, preprocessing, model design, and model implementation using Python programming. The training results of the backpropagation algorithm show an error value of 0.537829578 after 1,000 epochs, while the evaluation using Mean Squared Error (MSE) indicates an MSE of 0.022709 during the training process and 0.017604 during the testing process. The model also produces CPO price predictions for the next three months, namely 932.578 for the first month, 949.568 for the second month, and 774.855 for the third month. These findings indicate that the developed model is capable of predicting future CPO prices with adequate accuracy, which can assist companies in making better financial decisions and managing risks associated with CPO price fluctuations.

The increasing number of Indonesian Migrant Workers (TKW) working abroad, particularly through programs organized by BP2MI, has become a significant concern in managing the labor export process. One of the challenges faced is the uncertainty of the number of TKW to be sent each year, which is influenced by various external and internal factors. Therefore, this study aims to apply artificial neural networks (ANN) with a backpropagation algorithm approach to predict the number of TKW that will be processed by BP2MI. This method was chosen due to its ability to recognize patterns and nonlinear relationships between variables that affect the decision-making process for TKW export. In this study, the data used includes factors such as the number of job seekers, government policies, and the condition of the international labor market. The artificial neural network with the backpropagation algorithm is used to train the model based on existing historical data, with the goal of generating accurate predictions regarding the number of TKW to be processed in the coming years. The results of the tests show that the developed model can provide fairly accurate predictions and can serve as a tool for BP2MI in planning and managing the export of TKW more effectively. With the application of this technology, it is expected that the decision-making process related to TKW export can become more efficient and well-predicted.

This study aims to design, implement, and compare the performance of the Backpropagation algorithm from Artificial Neural Networks and the Seasonal Autoregressive Integrated Moving Average (SARIMA) model in predicting the optimal daily rice requirement at Grillme Restaurant in Pontianak. The main problem faced by the restaurant is the uncertainty in determining the required daily rice stock, which periodically results in either understocking (shortage) or overstocking (wastage), leading to operational losses. To address this, the study utilizes historical daily rice sales data from January 2023 to April 2025 as the database for training and testing both predictive models. The SARIMA approach is employed to capture time series components (trend and seasonality), while Backpropagation is utilized to model non-linear patterns. Comparative test results indicate that the SARIMA model achieved superior accuracy compared to the Backpropagation model. This is confirmed by the Mean Absolute Percentage Error (MAPE) value of the SARIMA algorithm being 17.35%, which is lower than the MAPE value of Backpropagation at 19.62%. The MAPE values obtained by both models demonstrate good predictive capability, but it is concluded that SARIMA is more recommended for a more efficient and planned management of rice stock at Grillme Restaurant in Pontianak.

This study develops a computational-based system to identify individual potential through the analysis of signature patterns using Artificial Neural Networks (ANN) and the Backpropagation algorithm. The research aims to explore and examine the effectiveness of applying ANN in recognizing and identifying signature patterns that are assumed to be related to an individual’s potential. In the data processing stage, Principal Component Analysis (PCA) is employed as a dimensionality reduction and feature extraction technique to optimally obtain the main characteristics of signature images. The system performance evaluation is conducted using a total of 80 signature images, consisting of 60 training data and 20 testing data. This study analyzes two network architecture configurations, namely a model with one hidden layer and a model with two hidden layers. The experimental results show that both network configurations achieve the same accuracy level of 92.5%. These findings indicate that the use of Artificial Neural Networks with the Backpropagation algorithm is effective in producing high accuracy in the signature pattern recognition process. Furthermore, the developed system has broad potential applications in the field of personal identification, such as employee evaluation, selection systems, and other applications across various organizational and industrial sectors.

This study explores the integration of deep learning based approaches in real time video content analysis for intelligent human computer interaction (HCI) in multimedia systems. Traditional video analysis techniques, such as rule-based methods and offline processing, struggle with real time performance and adaptability to complex video data. In contrast, the deep learning model used in this research, particularly Convolutional Neural Networks (CNNs), provides high accuracy in object detection, feature extraction, and real time processing. The integration of CNNs with interactive visualization modules enables dynamic adjustments to video content based on user interactions, ensuring a seamless and engaging user experience. The system was benchmarked in terms of its processing speed, accuracy, and responsiveness, showing significant improvements over traditional approaches in real time video analysis. Moreover, the study demonstrates that combining deep learning with real time visualization enhances the efficiency of interactive multimedia applications, making it suitable for dynamic environments such as surveillance, security monitoring, and interactive media. Despite the system's strong performance, challenges such as computational demands in high-resolution video processing were identified, highlighting the need for further optimization. Future work will focus on optimizing the system for different hardware platforms, incorporating multimodal inputs, and refining deep learning models to address computational bottlenecks. This research contributes to advancing HCI by providing insights into the integration of deep learning for real time video content analysis, which is pivotal for enhancing the interactivity and adaptability of intelligent multimedia systems.

The increasing adoption of collaborative robots in modern manufacturing environments requires reliable perception systems that can ensure both safety and operational efficiency during human–robot collaboration. This study proposes a CNN-based real-time computer vision system for object and human detection in shared robotic workspaces. The research focuses on developing and evaluating a single-stage deep learning detection model optimized for real-time performance while maintaining high detection accuracy. The proposed methodology includes dataset preparation, model training using transfer learning, real-time system implementation, and comprehensive performance evaluation. Experimental results demonstrate that the developed system achieves high detection accuracy, as reflected by strong precision, recall, and mean Average Precision (mAP) values, while maintaining low inference latency suitable for real-time operation. The system consistently operates above real-time frame-rate thresholds, ensuring timely perception updates required for safety-related decision-making in collaborative robotic environments. Graphical and quantitative analyses further confirm the stability of inference performance under dynamic interaction scenarios involving human movement and multiple objects. Compared with existing approaches, the proposed system provides a balanced trade-off between accuracy and computational efficiency, making it practical for deployment in safety-aware human–robot collaboration scenarios. Overall, the findings indicate that CNN-based real-time object detection systems can effectively support perception and situational awareness in collaborative robotics, contributing to safer and more efficient industrial automation.

Human–Robot Interaction (HRI) systems increasingly rely on data-driven approaches to interpret multimodal sensory inputs and support natural interaction. However, purely neural-based HRI models often suffer from limited interpretability and insufficient context-aware decision-making, which can reduce user trust and adaptability in dynamic interaction scenarios. To address these limitations, this study proposes a hybrid neural–symbolic HRI framework that integrates multimodal neural perception with explicit symbolic reasoning for adaptive and interpretable robot behavior. The proposed system combines deep neural networks for processing visual, speech, and gesture inputs with a rule-based symbolic reasoning layer that models interaction context, user states, and behavioral constraints. A loosely coupled integration strategy enables neural outputs to be transformed into symbolic representations, allowing logical inference to guide action selection while preserving perceptual accuracy. The framework was evaluated through controlled HRI experiments comparing a neural-only baseline with the proposed hybrid configuration across multiple interaction scenarios. Experimental results demonstrate that the hybrid neural–symbolic system significantly improves interaction accuracy, contextual responsiveness, and user satisfaction, while achieving substantial gains in interpretability. These findings indicate that symbolic reasoning effectively complements neural perception by enhancing transparency and context-aware adaptation without compromising performance. The study concludes that hybrid neural–symbolic architectures provide a promising foundation for developing trustworthy, adaptive, and human-centered HRI systems.

Explainable Artificial Intelligence (XAI) has become a critical area of research within artificial intelligence, focusing on improving the transparency and interpretability of machine learning (ML) models, often referred to as "black-box" models. The need for XAI techniques arises from the inherent complexity of ML models, which can make their decision-making processes difficult for users to understand. This study investigates various XAI techniques, including LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations), to assess their impact on model interpretability without significantly compromising predictive performance. A comparative experimental design was used, applying these XAI methods to different ML models, including deep neural networks and ensemble methods, within large-scale enterprise data analytics systems. The results indicate that XAI methods significantly enhance model transparency and decision traceability, allowing users to understand the influence of individual features on predictions. While a slight reduction in predictive accuracy was observed, especially with simpler models, the trade-off between interpretability and performance was deemed acceptable, particularly in fields requiring transparency, such as healthcare, finance, and autonomous systems. The use of XAI in enterprise data systems has practical implications for fostering trust and enabling informed decision-making among stakeholders. Furthermore, the study discusses the challenges and limitations of applying XAI techniques, such as complexity, scalability, and model-specific limitations. Future research is suggested to focus on developing more scalable and efficient XAI methods, enhancing their applicability across various model types, and addressing the challenges of real-time applications. This will be crucial in ensuring the widespread adoption of XAI in critical domains, promoting the ethical use of AI while maintaining predictive accuracy.

Traditional association rule analysis is effective at uncovering co-purchase patterns but fails to provide a global structural view of the market, which often results in fragmented and isolated insights. This study proposes a hybrid framework that integrates the Apriori algorithm with a Minimum Spanning Tree (MST) in order to validate and contextualize association rules within a single structural backbone. Transaction data from a retail store are transformed into a weighted, undirected product graph using an inverse-support function, and an MST is then extracted to represent the market backbone, while frequent itemsets and strong rules are obtained using Apriori. Experimental results on 236 multi-item transactions show that the MST backbone comprises 10 products and 9 fundamental links, with 66.67% of these links being confirmed by strong association rules, indicating a substantial coherence between statistical and structural evidence. The proposed model identifies 41 Apriori patterns that can be embedded in the MST and ranks them using a new metric, Structural Distance, which enables the categorization of Core Patterns, Bridge Patterns, and Complex Patterns according to their structural tightness. This hybrid perspective distinguishes dense, strategically meaningful bundles from anomalous but frequent combinations that are structurally peripheral, thereby offering a more holistic and actionable alternative to conventional Market Basket Analysis. The validated framework can support various applications, including store layout optimization, cross-selling strategies, and the design of path-based recommender systems, and it opens avenues for future extensions based on dynamic graphs and Graph Neural Networks.

Recommender systems play a critical role in shaping user decisions across digital platforms; however, the increasing complexity of recommendation algorithms has raised serious concerns regarding transparency, trust, and accountability. This study focuses on enhancing the transparency of recommender systems by integrating Explainable Artificial Intelligence (XAI) techniques within a MovieLens-based recommendation framework. The primary problem addressed is the opacity of conventional recommendation models, which limits user understanding of why certain items are recommended and may reduce trust, perceived fairness, and system acceptance. Accordingly, the main objective of this research is to design and evaluate a hybrid explainable recommender system that balances predictive accuracy with human-understandable explanations. The proposed approach combines Matrix Factorization, feature-importance-aware neural networks, and knowledge graph embeddings to construct a robust recommendation model. To enhance explainability, multiple XAI strategies are integrated, including model-agnostic methods (LIME, SHAP, and CLIME), argumentation-based explanations, and context-aware personalized explanations. A comprehensive evaluation framework is employed, incorporating algorithmic metrics (accuracy, fidelity, robustness, counterfactual consistency, and fairness) alongside human-centered evaluations measuring trust, transparency, cognitive load, and perceived usefulness. Experimental results demonstrate that the knowledge graph–enhanced hybrid model achieves superior recommendation accuracy compared to baseline approaches. Moreover, context-aware explanations consistently outperform other methods in terms of fidelity, robustness, and user-perceived transparency, while argumentation-based explanations are found to be the most persuasive. CLIME offers a strong balance between technical stability and interpretability. The findings indicate that no single explainability technique is universally optimal; instead, hybrid and adaptive explanation strategies are most effective. In conclusion, this study confirms that human-centered, context-adaptive XAI significantly improves transparency and user trust in recommender systems, highlighting explainability as a fundamental component rather than an optional enhancement.

End-to-end autonomous driving has emerged as a promising paradigm in which deep neural networks directly map raw visual inputs to continuous control actions. Despite its effectiveness, this approach suffers from limited transparency, posing significant challenges for deployment in safety-critical driving scenarios. This study addresses the lack of interpretability in vision-based end-to-end autonomous driving systems and aims to analyze model decision-making behavior under critical conditions such as sharp steering maneuvers and abrupt control transitions. To this end, an explainable end-to-end autonomous driving framework is proposed, combining a convolutional neural network trained via imitation learning with gradient-based visual attribution techniques, including Grad-CAM. The model predicts continuous steering, throttle, and braking commands directly from front-facing camera images, while explainability mechanisms are applied to reveal input regions influencing each control decision. Model performance is evaluated using both prediction accuracy and safety-oriented behavioral metrics. Experimental results show that the proposed explainable model achieves lower control prediction errors compared to a baseline end-to-end CNN, reducing steering mean squared error from 0.034 to 0.031, throttle error from 0.021 to 0.019, and brake error from 0.018 to 0.016. Moreover, safety-oriented analysis indicates improved driving stability, with steering variance reduced from 0.087 to 0.072 and abrupt control changes decreased from 14.6 to 10.3 events. Visual explanations consistently highlight road surfaces and lane-related structures during complex maneuvers, indicating reliance on semantically meaningful cues. In conclusion, the results demonstrate that integrating explainability into end-to-end autonomous driving not only preserves predictive performance but also correlates with smoother and more stable driving behavior. This framework contributes to the development of transparent and trustworthy autonomous driving systems suitable for safety-critical applications

This study discusses the introduction of digital signature patterns using the Backpropagation method on Artificial Neural Network (JST) to identify a person's characteristics and potential. The increasing use of digital identities demands a verification system that is more secure, accurate, and adaptive to the variations of each individual's signature. The main problem faced in the signature recognition system is the low level of accuracy when the visual features of the signature have similarities between users, both in terms of shape, size, and stroke pressure. In addition, variations of signatures made by the same individual are also a challenge in the identification process. As a solution, this study implements Principal Component Analysis (PCA) to extract important features from the signature image before the training process using JST. PCA is used to reduce the data dimension so that the learning process becomes more efficient and optimal. A total of 80 signature images were used in this study, consisting of 60 training data and 20 test data. The results showed that the system was able to achieve an accuracy level of 92.5%. These findings prove that the combination of PCA and JST methods is effective in recognizing digital signature patterns and has the potential to be applied to digital security-based biometric identification systems.

This study explores the intersection of Artificial Intelligence (AI), agriculture, and decolonial philosophy, emphasizing the role of local knowledge as the foundation for sustainable agricultural technology in Indonesia. The research investigates how AI can be developed not as a tool of technological domination but as a dialogical partner that recognizes the epistemic value of indigenous wisdom. Using a mixed-method approach, the study combines algorithmic experiments applying lightweight Convolutional Neural Networks (CNN) with Explainable AI (XAI) methods such as SHAP and LIME with participatory interviews involving farmers in Bima District. Empirical findings show that models integrated with localized visualization and community-based interpretability improved user trust by 84% and reduced computational energy by 28% without compromising accuracy. More importantly, the interaction between AI and farmers revealed a form of epistemic integration where algorithmic logic aligns with traditional indicators, such as soil texture, humidity, and seasonal signs known to local farmers. Philosophically, this research asserts that sustainable AI should emerge from ecological and cultural contexts rather than imposing external frameworks. In the decolonial sense, it positions local farmers not as passive users but as active epistemic agents shaping the meaning of technology. Thus, AI becomes not only a technical instrument but a site of ethical and epistemic liberation that reaffirms human responsibility toward knowledge, culture, and the earth.

The rapid growth of open-source software (OSS) in machine learning (ML) has intensified the need for reliable, automated methods to assess project quality, particularly as OSS increasingly underpins critical applications in science, industry, and public infrastructure. This study evaluates the effectiveness of a diverse set of machine learning and deep learning (ML/DL) algorithms for classifying GitHub OSS ML projects as engineered or non-engineered using a SMOTE-enhanced and explainable modeling pipeline. The dataset used in this research includes both numerical and categorical attributes representing documentation, testing, architecture, community engagement, popularity, and repository activity. After handling missing values, standardizing numerical features, encoding categorical variables, and addressing the inherent class imbalance using the Synthetic Minority Oversampling Technique (SMOTE), seven different classifiers—K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), XGBoost (XGB), Logistic Regression (LR), Support Vector Machine (SVM), and a Deep Neural Network (DNN)—were trained and evaluated. Results show that LR (84%) and DNN (85%) outperform all other models, indicating that both linear and moderately deep non-linear architectures can effectively capture key quality indicators in OSS ML projects. Additional explainability analysis using SHAP reveals consistent feature importance across models, with documentation quality, unit testing practices, architectural clarity, and repository dynamics emerging as the strongest predictors. These findings demonstrate that automated, explainable ML/DL-based quality assessment is both feasible and effective, offering a practical pathway for improving OSS sustainability, guiding contributor decisions, and enhancing trust in ML-based systems that depend on open-source components.

Application of Multi-Layer Perceptron neural network to fault classification in high-voltage transmission lines is demonstrated in this paper. Different fault types on protected transmission line should be detected and classified rapidly and correctly. This paper presents the use of Discrete Wavelet Transform energy features combined with zero sequence current magnitude as input features for neural network classifier. The proposed method uses eight extracted features to learn hidden relationship in fault signal patterns. Using proposed approach, fault detection and classification of all 11 fault types could be achieved with high accuracy. Improved performance is experienced once the neural network is trained sufficiently with 1188 fault samples, thus performing correctly when faced with different system conditions. Results of performance studies show that proposed neural network-based classifier achieves 96.18% average accuracy, which demonstrates that it can improve the performance of conventional fault classification algorithms, which in turn can provide more efficient solutions in the management and protection of high voltage electrical systems.

Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disorder that presents significant diagnostic challenges due to its heterogeneous clinical manifestations and symptom overlap with other neurological conditions. Early and accurate diagnosis is critical for initiating timely interventions and improving patient outcomes. Traditional diagnostic approaches rely heavily on clinical expertise and manual interpretation of neuroimaging data, such as structural MRI, Diffusion Tensor Imaging (DTI), and functional MRI (fMRI), which are inherently time-consuming and prone to interobserver variability. Recent advances in Artificial Intelligence (AI) and Deep Learning (DL) have demonstrated potential for automating neuroimaging analysis, yet existing models often suffer from limited generalizability across modalities and datasets. To address these limitations, we propose a Transformer-augmented deep learning ensemble framework for automated ALS diagnosis using multi-modal neuroimaging data. The proposed architecture integrates Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Vision Transformers (ViTs) to leverage the complementary strengths of spatial, temporal, and global contextual feature representations. An adaptive weighting-based fusion mechanism dynamically integrates modality-specific outputs, enhancing the robustness and reliability of the final diagnosis. Comprehensive preprocessing steps, including intensity normalization, motion correction, and modality-specific data augmentation, are employed to ensure cross-modality consistency. Evaluation using 5-fold cross-validation on a curated multi-modal ALS neuroimaging dataset demon-strates the superior performance of the proposed model, achieving a mean classification accuracy of 94.5% ± 0.7%, precision of 93.9% ± 0.8%, recall of 92.9% ± 0.9%, F1-score of 93.4% ± 0.7%, spec-ificity of 97.4% ± 0.6%, and AUC-ROC of 0.968 ± 0.004. These results significantly outperform baseline CNN models and highlight the potential of transformer-augmented ensembles in complex neurodiagnostic applications. This framework offers a promising tool for clinicians, supporting early and precise ALS detection and enabling more personalized and effective patient management strategies.