Publication Search

This study aims to analyze the effect of artificial intelligence (AI)-based demand forecasting and big data analytics on inventory optimization through prediction accuracy among e-commerce actors or marketplace sellers in Curug. This research employed an explanatory quantitative approach involving 100 respondents selected through purposive sampling based on predefined criteria relevant to the research objectives. Data were collected using a structured Likert-scale questionnaire and analyzed using Structural Equation Modeling Partial Least Squares (SEM-PLS) by evaluating the measurement model, structural model, and mediation effects. The findings reveal that AI-based demand forecasting and big data analytics have a positive and significant effect on prediction accuracy. Furthermore, prediction accuracy has a positive and significant effect on inventory optimization and significantly mediates the relationship between AI-based demand forecasting, big data analytics, and inventory optimization. These findings indicate that the integration of AI and big data analytics contributes to more accurate demand prediction, leading to improved inventory management performance. The implication of this study suggests that e-commerce actors should improve data quality, analytical capability, and the adoption of predictive technologies to support more accurate, efficient, and responsive inventory decisions, thereby enhancing operational performance and competitiveness in responding to dynamic market demand.      

Evaluating food portions and types in the Free Nutritious Meal (MBG) program is generally still performed manually, which is time-consuming and potentially subjective. This study aims to develop an automated deep learning-based system to efficiently detect food types and estimate their nutritional value. The method used is a quantitative experiment integrating the YOLOv11m architecture for real-time object detection and the Google Gemini 2.5 Flash Large Language Model (LLM) for contextual nutritional estimation reasoning. The model training utilized a dataset of 2,630 food tray images categorized into five classes (fruit, side dish, staple food, vegetable, milk) that had undergone an augmentation process. The results showed that the YOLOv11m model achieved excellent performance with a mean Average Precision (mAP@0.5) of 0.9727 and the highest F1-score of 0.9522 at a confidence threshold of 0.1. Furthermore, validation of the LLM integration demonstrated a high prediction agreement rate of 85%. In conclusion, the combination of the YOLOv11m algorithm and LLM reasoning is capable of detecting and validating nutritional classification quickly and precisely, showing strong potential as an objective nutritional evaluation monitoring solution for large-scale MBG program implementation.

This study investigates the integration of digital twin technology and machine learning for predictive analysis in smart mechanical systems. The research emphasizes the role of intelligent computational frameworks in improving industrial monitoring, predictive maintenance, and operational efficiency within Industry 4.0 environments. A qualitative content analysis approach was employed by reviewing scientific literature, industrial reports, and previous studies related to digital twins, artificial intelligence, and predictive analytics. The findings indicate that digital twin architectures supported by machine learning algorithms can significantly enhance real-time monitoring, fault prediction accuracy, and maintenance optimization. The integration of IoT devices, cloud computing, and intelligent analytics also improves industrial sustainability, reduces operational downtime, and supports data-driven decision-making processes. Furthermore, the study identifies several technological challenges, including cybersecurity risks, data integration complexity, and computational limitations. Overall, the proposed intelligent digital twin framework provides a promising approach for future industrial innovation and sustainable smart mechanical system management

Stunting is a multifactorial public health problem influenced by various risk factors that may emerge during the prenatal period. Early identification of stunting risk during pregnancy is important to support preventive interventions. This study aimed to develop a stunting risk prediction model based on maternal prenatal factors using the Random Forest algorithm. Secondary data from 172 pregnant women, consisting of 83 stunting cases and 89 non-stunting cases, were analyzed. The predictor variables included maternal age during pregnancy, height, hemoglobin level, mid-upper arm circumference (MUAC), smoking history, hypertension, asthma, and diabetes mellitus. The research stages consisted of data preprocessing, model training using Stratified 5-Fold Cross Validation, performance evaluation, external testing, and feature importance analysis. Internal evaluation results showed an accuracy of 60%, precision of 60.6%, recall of 57.3%, F1-score of 58.9%, and AUC of 0.6688. External testing yielded an accuracy of 70% and an AUC of 0.6167. Feature importance analysis identified maternal age during pregnancy as the most influential variable in the prediction process. The findings indicate that maternal prenatal factors have potential for early stunting risk identification, although the predictive performance remains moderate. This approach may serve as a foundation for developing early screening tools to support targeted interventions among high-risk pregnancies.

Bank stock price prediction is an important topic in the application of information technology because stock price movements are dynamic, sequential, and influenced by historical market patterns. This study aims to predict Indonesian banking stock prices using the Long Short-Term Memory method and evaluate the effect of Bayesian Optimization on model performance. The data used in this study consists of daily historical stock data of BBCA, BBNI, BBRI, BBTN, and BMRI from May 4, 2020, to May 4, 2026, obtained from Yahoo Finance. The input features include opening price, highest price, lowest price, closing price, and trading volume, while the prediction target is the stock closing price. The results show that the baseline model produced MAPE values ranging from 1.892% to 3.147%. The best baseline performance was obtained on BBCA with an R² value of 0.933, followed by BBTN with an R² value of 0.902. After optimization, performance improvement occurred on BBTN, with MAPE decreasing from 3.147% to 2.482% and R² increasing from 0.902 to 0.935. For BMRI, MAPE decreased from 2.385% to 2.206%, and R² increased from 0.687 to 0.743. This study concludes that Long Short-Term Memory can be used to predict Indonesian banking stock prices, while Bayesian Optimization can selectively improve model performance depending on the characteristics of each stock dataset.

Motor Vehicle Tax (PKB) is a key pillar of Regional Original Revenue (PAD) that supports development funding. However, seasonal fluctuations in payment realization create uncertainties in local budget planning. This study aims to address the limitations of the standard Random Forestalgorithm, which suffers from extreme prediction failures on time-series data due to its inability to capture temporal transitions between months. The proposed solution implements feature engineering using a Cyclical Encoding approach (Sine and Cosine transformations) and Lagged Variables. The dataset comprises historical records of motor vehicle tax potential and realization from January 2021 to November 2025. The baseline model evaluation without feature engineering yields highly inaccurate predictions with a Mean Absolute Percentage Error (MAPE) of 203.47% (accuracy of -103.47%). Conversely, after integrating Cyclical Encoding and Lagged Variables, the proposed model's performance improves drastically, achieving a MAPE of 14.40% (an accuracy rate of 85.60%), an MAE of 9,317 units, and an RMSE of 12,638 units. Feature Importance analysis confirms that the cyclically encoded month feature contributes the highest weight to the model's decisions with a score of 0.5031, followed by the potential feature at 0.1798. This study demonstrates that time-based feature engineering effectively optimizes Random Forestfor precise tax revenue forecasting.

Early detection of a potential heart attack is a crucial step in preventing sudden death from heart disease. This research aims to develop an Internet of Things (IoT)-based health monitoring system capable of measuring vital body data in real time and predicting the likelihood of a heart attack from CSV data obtained from sensors, integrated through RapidMiner as learning data using a machine learning algorithm, the Support Vector Machine (SVM). The system was built using an ESP32 microcontroller connected to a MAX30102 sensor to measure heart rate and finger oxygen levels (SpO₂), as well as a DHT22 sensor to measure temperature and humidity. The resulting data is sent to the Blynk application to display real-time data according to its parameters. The initial prediction logic was developed using a rule-based method based on medical thresholds for four vital parameters. The data was then used to train an SVM model as a classification system to detect potential heart attacks. Test results showed that the system can identify abnormal conditions with a good level of accuracy and provide early warnings based on changes in vital parameters in real time. This system is expected to be an initial solution for personal health monitoring, especially for individuals at risk of heart disease. It can be further developed with cloud integration and automatic notifications to users' devices.

The increasing use of mobile applications in the retail industry has generated a large volume of user reviews that contain valuable insights regarding customer experience and service quality. However, the unstructured nature of these reviews requires an automated approach to extract meaningful patterns efficiently. This study aims to perform sentiment analysis on user reviews of the Indomaret Poinku application by integrating lexicon-based labeling with machine learning classification. A total of 10,000 reviews were collected from Google Play Store and processed through a series of text preprocessing steps, including cleaning, case folding, normalization, tokenization, stopword removal, and stemming. Sentiment labeling was performed using the Indonesian Sentiment Lexicon (InSet), producing three sentiment classes: positive, negative, and neutral. The labeled data were vectorized using CountVectorizer and classified using two algorithms: K-Nearest Neighbors (KNN) and Random Forest (RF). Evaluation results show that Random Forest outperforms KNN, achieving an accuracy of 82.5%, compared to 69% for KNN. Random Forest demonstrates superior performance in handling high-dimensional sparse text features and yields more stable predictions across sentiment classes. This study contributes to the growing body of research on Indonesian sentiment analysis by demonstrating the effectiveness of combining lexicon-based labeling with ensemble learning methods, offering practical implications for developers seeking to improve the quality and user satisfaction of digital retail applications.

Student dropout in higher education remains a persistent socioeconomic challenge, yet many predictive models reported in the literature are methodologically compromised by randomized cross-validation schemes that introduce temporal data leakage and artificially inflate predictive performance. This study proposes a longitudinal prescriptive learning analytics framework integrating three complementary methodological components: a Leave-One-Cohort-Out (LOCO) temporal validation protocol, a hybrid SMOTE-ENN class balancing strategy, and temporal velocity feature engineering derived from Learning Management System (LMS) behavioral trajectories. The framework was evaluated on a longitudinal dataset comprising 464,739 enrollment records and 77 features. Five predictive algorithms—XGBoost, LightGBM, CatBoost, Random Forest, and Logistic Regression—were comparatively assessed on a strictly isolated blind holdout cohort (2022), with CatBoost emerging as the champion estimator, achieving a PR-AUC of 0.8859, a Macro F1-Score of 0.9143, and the lowest Brier Score (0.0221), thereby demonstrating superior calibration and discriminative capability under severe class imbalance (93:7 ratio). Comprehensive ablation analysis revealed that temporal velocity features function not merely as additive predictors, but as a structural prerequisite enabling Synthetic Minority Oversampling Technique with Edited Nearest Neighbors (SMOTE-ENN) to generate high-quality synthetic boundary instances; removing these features reduced minority-class precision from 0.8302 to 0.6721. To operationalize predictive outputs into actionable intervention pathways, Diverse Counterfactual Explanations (DiCE) were implemented under a three-tier causal constraint architecture on 96 borderline high-risk students, generating 384 feasible intervention scenarios exclusively targeting forward-looking behavioral velocity metrics without constraint violations. Collectively, these findings advance the paradigm of prescriptive learning analytics by providing educational institutions with interpretable risk diagnostics and operationally feasible intervention guidance grounded in empirically validated behavioral and temporal dynamics.

 Cryptocurrency price prediction, particularly for highly volatile assets like Solana (SOL), is a crucial challenge in time series data analysis in digital finance. This study aims to compare the performance of the XGBoost machine learning algorithm with the Temporal Fusion Transformer (TFT) deep learning model in predicting Solana's daily closing price. The dataset used consists of historical Solana price data and network fundamentals data in the form of Total Value Locked (TVL). The research process includes data preprocessing, dividing training and test data, model training, and evaluation using the Root Mean Squared Error (RMSE) metric. The results show that using the same-day price feature has the potential to cause target leakage, resulting in invalid prediction accuracy. In testing using pure historical data without data leakage, the XGBoost model performed better than TFT with an RMSE of 4.27, while TFT produced an RMSE of 18.59. Furthermore, the integration of network fundamentals data in the form of TVL did not improve prediction accuracy and even caused a decrease in performance for the XGBoost model with an RMSE of 7.10. The results of this study show that the use of historical price action features is more effective than fundamental network indicators for short-term daily Solana price predictions.

Blasting is a critical method for overburden removal in open-pit coal mines, where fragmentation quality directly impacts loading efficiency and operational costs. This study aims to analyze the actual fragmentation resulting from overburden blasting at Pit 4 Middle of PT. Victor Dua Tiga Mega, Central Kalimantan, to predict fragmentation using the Kuz-Ram model, and to evaluate the conformity of both results against the company standard (boulder size ≤144.6 cm or ≥50 cm for analysis). The research employed a quantitative comparative method. Primary data included blasting geometry and photographs of muck piles from 10 blasting events, which were analyzed using WipFrag software to obtain actual fragmentation distribution. Secondary data comprised rock characteristics and explosive properties for Kuz-Ram prediction input. The results showed significant variation in actual boulder percentage (≥50 cm), ranging from 6.19% to 32.91% with an average of 16.05% (medium category). Statistical analysis revealed a very weak negative correlation (r = -0.21) between powder factor (PF) and boulder percentage, indicating that PF is not the dominant factor within the consistent application range (0.21-0.23 kg/bcm). Comparison with Kuz-Ram predictions showed that the model consistently over-predicted coarse material, with an average difference of +25.21%, suggesting the need for rock factor (A) recalibration. It is concluded that the blasting results are inconsistent, strongly influenced by uncontrollable factors such as geological conditions. Recommendations include geometri evaluation, particularly burden and spacing, and calibration of the Kuz-Ram model for more accurate future predictions.

Sentiment analysis for Indonesian regional languages faces two persistent challenges: labeled training data is extremely limited for most regional varieties, and transformer models pre-trained on Bahasa Indonesia do not generalize reliably to languages with substantially different morphological structures. Prior work on the NusaX benchmark has primarily relied on direct fine-tuning, treating each regional language independently and without exploiting linguistic proximity between related languages as a transfer signal. This paper proposes Language-Similarity-Guided Transfer (LSGT), a sequential fine-tuning strategy that first adapts a pre-trained model to a pivot language selected using character trigram similarity, followed by fine-tuning on the target language. Four transformer models are evaluated across all 12 NusaX languages using the official train/validation/test splits: IndoBERT, NusaBERT, mBERT, and XLM-R. Performance is evaluated using four metrics: accuracy, macro F1, macro precision, and macro recall. Experimental results show that LSGT improves macro F1 in 44 of 48 model-language combinations, demonstrating that the fine-tuning strategy itself is a major factor in low-resource cross-lingual sentiment classification. XLM-R benefits most strongly from LSGT, achieving an average improvement of +0.137 macro F1 and a peak gain of +0.298 on Madurese. SHAP-based token attribution analysis further reveals that predictions rely heavily on named entities and domain-specific nouns rather than sentiment-bearing vocabulary, indicating a dataset-level bias inherited from the original SmSA corpus and propagated through the NusaX translation pipeline.

This study analyzes the performance of three types of controllers—PID, Fuzzy-PID, and Model Predictive Control (MPC)—in maintaining water level stability in a dam system. Proper water level regulation is essential to prevent flooding, ensure adequate water distribution, and maintain the structural safety of the dam. Previous studies have widely implemented PID and Fuzzy-PID controllers; however, MPC offers significant advantages through system prediction and multivariable optimization. This research compares the performance of the three controllers through simulation analysis using a second-order mathematical model. The results indicate that MPC has the potential to deliver a faster settling time and a smaller steady-state error compared to PID and Fuzzy-PID, although it requires higher computational complexity.

This study examines customer churn prediction in subscription-based telecommunications from a digital marketing perspective using machine learning. The analysis utilizes a secondary dataset of 7,043 customer records that simulate behavioral, contractual, and financial attributes commonly found in telecom services. Three classification algorithms Logistic Regression, Random Forest, and Gradient Boosting are applied to model churn behavior. Data preprocessing includes handling missing values, encoding categorical variables, and splitting data into training and testing sets. Model performance is evaluated using accuracy, recall, and ROC-AUC, with emphasis on recall due to its importance in identifying at-risk customers. The results show that Gradient Boosting achieves the highest overall performance with an ROC-AUC of 0.84, while Logistic Regression provides relatively higher recall. Key drivers of churn include short-term contracts, higher monthly charges, and lower service engagement. However, recall remains moderate, indicating limitations in capturing complex behavioral factors. These findings suggest the need to combine predictive models with behavioral insights and highlight the importance of early customer engagement and long-term retention strategies.

The rapid development of Artificial Intelligence (AI) has brought significant changes to the criminal justice system, particularly in criminal investigations and evidentiary processes, while simultaneously raising complex legal and ethical challenges. Objective: This study aims to analyze the legal implications of the use of AI in criminal investigations, focusing on its benefits, risks, and challenges related to the admissibility of AI-based evidence, as well as the need for regulatory frameworks that ensure fairness, transparency, and accountability. Methods: This research employs a normative qualitative approach through the analysis of legal regulations, a review of legal and technological literature, and a comparative approach across jurisdictions, complemented by case studies of AI applications in law enforcement practices. Results: The findings indicate that AI enhances investigative efficiency through data analysis, crime prediction, and digital forensics; however, it also poses risks such as algorithmic bias, human rights violations, and issues concerning the reliability and transparency of evidence. Furthermore, differences across legal systems result in the absence of uniform standards for the admissibility of AI-based evidence. Therefore, adaptive regulatory frameworks grounded in the principles of fairness, transparency, and accountability are required, along with strengthened human oversight to ensure that the use of AI aligns with the principles of justice and human rights protection.

Introduction: Additive Manufacturing (AM) has revolutionized the production of complex geometries, offering flexibility, customization, and precision across various industries. However, optimizing multiple process parameters simultaneously to enhance AM performance remains a significant challenge. This study focuses on improving both mechanical properties and surface quality by utilizing multi-objective optimization techniques. Literature Review: The research reviews existing approaches in AM optimization, highlighting the limitations of single-objective optimization and the potential of multi-objective evolutionary algorithms (MOEAs). Previous studies demonstrate the difficulty of balancing competing objectives, such as tensile strength and surface roughness, within AM processes. Materials and Method: This study employs NSGA-II, MOEA/D, and SPEA2 algorithms to optimize AM parameters like layer thickness, build orientation, and infill density. The optimization aims to improve mechanical performance, including tensile strength and impact resistance, while reducing build time and surface roughness. The methodology integrates experimental validation with computational predictions to evaluate the effectiveness of these algorithms. Results and Discussion: The optimization process yielded Pareto-optimal solutions that balanced mechanical strength and surface quality. The results demonstrated improvements in tensile strength and surface finish without significantly increasing build time. Trade-off analysis highlighted the inherent conflicts between mechanical performance and surface quality, allowing for better decision-making in industrial applications. The study contributes to the AM industry by offering a comprehensive optimization framework for improving both efficiency and product quality.

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.

Software Defect Prediction (SDP) aims to identify defective modules early in the software development lifecycle to improve software quality and reduce maintenance costs. However, SDP datasets commonly suffer from high dimensionality, feature redundancy, and class imbalance, which can degrade model performance and stability. This study proposes a hybrid feature selection framework to address these challenges and enhance prediction performance. The proposed approach integrates Combined Correlation and Mutual Information (CONMI), which combines the Pearson Correlation Coefficient (PCC) and Mutual Information (MI) to capture both linear and nonlinear feature relevance. The selected features are further refined through Top-K selection, correlation-based filtering to reduce multicollinearity, and Backward Elimination (BE) to obtain an optimal feature subset. To address class imbalance, SMOTE-Tomek is applied by combining over-sampling and data cleaning techniques. Experiments are conducted on twelve NASA MDP datasets using Logistic Regression (LR) and Naïve Bayes (NB) classifiers. The results show that the proposed framework consistently achieves the best performance, with Logistic Regression combined with SMOTE-Tomek obtaining the highest average AUC of 0.7923 ± 0.0714, while NB achieves 0.7554 ± 0.0580. Statistical analysis using a paired t-test indicates that the proposed method significantly outperforms MI+SMOTE-Tomek and BE+SMOTE-Tomek for Logistic Regression, whereas no significant differences are observed for NB. In addition to improving overall classification performance (AUC), the proposed approach also enhances minority class detection, as reflected in improved Recall and F1-score. Overall, the proposed hybrid framework provides an effective and reliable solution for software defect prediction, particularly for high-dimensional and imbalanced datasets.

Cloud job-length prediction remains challenging when the target distribution is highly skewed and contains rare extreme values. This study proposes a log-transformed, regime-based machine learning framework for robust prediction of cloud job length, represented in million instructions (MI). The approach integrates sequential feature engineering, logarithmic target transformation, weighted learning, and regime-aware modeling to distinguish between normal and extreme job-length behavior. Using an ordered GoCJ-derived cloud job-length sequence of 1000 jobs, the dataset exhibits a heavy-tailed distribution, with a mean of 129,662 MI, a median of 93,000 MI, a 95th percentile of 525,000 MI, a 99th percentile of 900,000 MI, and a skewness of 3.695. The proposed model is evaluated against sequential baselines and stronger machine learning baselines, including Naive_Last, RollingMean_5, Global_Log_ExtraTrees, RandomForest, GradientBoosting, and MLP_Log. On the main test split, the proposed Regime_Log_ExtraTrees achieved the best RMSE of 206,255.66 and the least negative R² of −0.01062, while Global_Log_ExtraTrees remained competitive in terms of MAE, MedAE, and RMSLE. Additional walk-forward validation confirms that the regime-aware model consistently achieves the best mean RMSE and mean R² across temporal folds. Ablation results further show that regime-aware learning is the primary contributor to robustness, although accurate prediction of extreme jobs remains challenging. These findings indicate that log-transformed, regime-based learning provides a practical and more robust strategy for cloud job-length prediction under heavy-tailed workload conditions.

This study aims to analyze the phytochemical profile of Sambung Nyawa (Gynura procumbens) leaves as a potential herbal candidate for mild hypertension therapy using in silico methods. Plant samples were examined to identify active compounds documented in the PubChem database. The identified compounds were further analyzed using PASS Online to predict their pharmacological activities, ProTox-II to evaluate toxicity levels, and pkCSM to assess ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics. The findings reveal that several bioactive compounds present in Sambung Nyawa leaves demonstrate strong predicted anti-hypertensive activity accompanied by minimal toxicological risk. PASS Online analysis indicates potential mechanisms of action, including vascular receptor modulation and mild diuretic properties that may support blood pressure regulation. ProTox-II classification places most compounds in the low-toxicity category, while pkCSM predictions confirm acceptable bioavailability and favorable pharmacokinetic properties. Overall, these results provide a preliminary scientific foundation for the development of Gynura procumbens as an alternative herbal therapy for mild hypertension and support the need for further validation through in vitro and in vivo experimental studies.