Vol. 5 No. 2 (2026): December

Aldo Grocery Store is a staple food retailer that still manually records inventory and lacks an integrated security system. This operational limitation leads to frequent recording errors, delays in restocking information, and vulnerabilities to theft. This research aims to develop a web-based inventory monitoring and security system integrated with an Arduino Uno microcontroller. The system utilizes ultrasonic sensors to monitor item availability on shelves based on distance, load cell sensors to track stock weight, along with Passive Infrared (PIR) sensors and an RFID module to detect unauthorized movement and verify access identification. Data collected from all sensors is processed by the Arduino Uno and transmitted to a server database via the ESP8266 WiFi module, presenting updates on a web interface in real time. Test results indicate that the system measures commodity weight via the load cell sensor with an accuracy of 95%, responds to security triggers with an average response time of less than 2 seconds, and reliably transmits data to the web server. This system facilitates store owners in managing inventory and enhances retail security management efficiently.

At Warkop Gasskeun East Jakarta, the determination of the best-selling menu item currently relies on the owner's intuition and manual observation, making the decision-making process subjective and unable to evaluate multiple performance criteria simultaneously. This study aims to implement the Simple Additive Weighting (SAW) method to determine the best-selling menu objectively through a decision support system framework. Evaluation was conducted across four criteria with predetermined weights: sales volume (50%), customer ratings (30%), profit (15%), and menu price (5%), categorized into benefit and cost attributes. Research data were gathered through direct observation, interviews with the business owner, and sales documentation, covering all available food and beverage items through total population sampling. The SAW procedure involved criteria weighting, decision matrix normalization, preference value calculation, and alternative ranking. The calculation results demonstrated that Indomie Double Telur achieved the highest preference score of 0.9667, establishing it as the primary best-selling menu item, closely followed by Nasi Daun Jeruk Telur with a score of 0.9625. The method was successfully integrated into a web-based decision support system, where verification confirmed an exact match between manual calculations and automated system outputs. Furthermore, black box testing confirmed that all functional modules—including data management, SAW processing, and reporting—operated as intended, thereby providing the business owner with a faster, objective, and data-driven decision-making tool. 

Coffee micro, small, and medium enterprises (MSMEs) require integrated inventory and financial management to ensure that raw-material purchasing decisions align with operational demand and budget constraints. Manual recordkeeping may result in delayed purchases, excess inventory, or stockouts. This article presents the design of a Multimodal Artificial Intelligence Financial Assistant to support automated purchasing recommendations for Kopi Menceng MSME. The proposed system is designed to accept text-based inputs and, as an optional feature, transaction data extracted from receipt images or documents. Validated data are processed alongside inventory levels, material requirements, prices, and budget limits to generate recommended purchase quantities and estimated costs. The development follows a prototyping approach comprising requirements identification, design, prototype construction, evaluation, and refinement. Proposed evaluation methods include black-box testing, recommendation-error measurement, and user assessment. Since actual implementation data have not been included, numerical examples in this article serve as simulations rather than field-test results. This design is intended to provide a foundation for developing a more structured purchasing decision-support system for MSMEs.

Fragmented inventory management in multi-warehouse environments causes recording errors, delayed stock synchronisation, and weak accountability. PT Satellite of Glow manages four separate warehouses relying on physical logbooks and isolated spreadsheets without a traceable audit trail. This study models and optimizes the company's inventory processes using Business Process Model and Notation (BPMN). A qualitative single-case study was conducted through direct observation and semi-structured interviews with four actor groups across the four warehouses. The workflow was modelled As-Is, evaluated using process redesign heuristics, and restructured into an automated To-Be model. Process performance was assessed by measuring the cycle times of ten high-frequency activities. The As-Is model revealed 18 manual activities, 12 of which were non-value-adding administrative tasks requiring identical data to be recorded three times. Redesigning the process via task elimination, automation, and database centralisation reduced the total cycle time from 150 to 30.5 minutes, achieving an efficiency gain of 79.67% with activity improvements ranging from 66.70% to 90.00%. Information-oriented activities attained higher efficiency (84.48%) than physical transactions (73.02%), confirming that physical handling defines the upper boundary of software-driven optimization. The To-Be model eliminates multi-tier recording and establishes automated audit-trail tracking.

The purpose of this study is to analyze and describe the program evaluation of the Online Licensing Management Information System (SIMPONIE) and identify its inhibiting factors in the environmental permit issuance process in South Tangerang City. The analytical framework applies Dunn's public policy evaluation model, encompassing effectiveness, efficiency, adequacy, equity, responsiveness, and accuracy. This study employed a descriptive qualitative method, collecting empirical data through interviews with informants from the Investment and Integrated One-Stop Services Agency (DPMPTSP) of South Tangerang City, the Environmental Agency (DLH) of South Tangerang City, and community permit applicants, analyzed using the Miles and Huberman interactive technique. The results indicate that the implementation of SIMPONIE in the environmental permit issuance process has generally performed adequately, although substantial improvements are still required in efficiency, adequacy, responsiveness, and accuracy. Inhibiting factors on the part of the implementer comprise network disruptions, constrained system and operator capacity, regulatory shifts, and inter-agency coordination challenges, whereas constraints on the applicant side stem from limited understanding of the electronic system, incomplete documentation, disparities in information technology literacy, and the persistence of in-person administrative habits. 

Forest and land fires are recurrent disasters in Kalimantan that have significant impacts on the environment, public health, and the economy. This study develops a hybrid model that combines Analytic Hierarchy Process (AHP) and Random Forest weighting using four criteria: rainfall, peatland, hotspot history, and distance to settlements. The final model assigns the highest weight to hotspot history (43.0%) as the strongest predictor. Four districts/cities, namely Ketapang, Pulang Pisau, Kubu Raya, and Palangka Raya, are classified as high risk (7.3% of the total evaluated districts/cities), while 26 districts/cities (47.3%) are classified as medium risk and 25 (45.5%) as low risk. Validation against 15,272 independent hotspot observations demonstrates a strong Pearson correlation (r = 0.718), a moderate Spearman correlation (ρ = 0.573), and significant differences among risk classes based on one-way ANOVA (F = 27.724, p < 0.0001). A ±20% weight sensitivity analysis indicates that the model is highly robust, while classification accuracy yields a Cohen’s Kappa of 0.288. Overall, the hybrid AHP–Random Forest model provides adequate validity to support decision-making in forest and land fire mitigation.

Twitter is one of the most widely used social media platforms in Indonesia, and telecommunication companies such as PT Telekomunikasi Seluler (Telkomsel) utilize their official account, @Telkomsel, as a primary channel for customer interaction and complaint handling. Currently, incoming complaint tweets are still classified manually by the customer service division, a procedure that is time-consuming and prone to human inconsistency. This study implements a text mining approach integrating the Naïve Bayes algorithm, Adaptive Boosting (AdaBoost), and the Synthetic Minority Over-Sampling Technique (SMOTE) to automatically classify Indonesian-language tweets mentioning @Telkomsel into two classes: "Complaint" and "Non-Complaint." A dataset of 2,000 tweets was collected through crawling in RapidMiner, followed by a two-stage text preprocessing workflow utilizing Gataframework and RapidMiner. The classification model was validated using 10-fold cross-validation. The experimental results demonstrate that the combination of Naïve Bayes, AdaBoost, and SMOTE achieves an accuracy of 78.84%, precision of 79.90%, recall of 80.51%, and an AUC of 0.809. These findings confirm that the proposed framework is capable of effectively automating complaint detection to support more responsive customer service operations.

Fast and accurate information retrieval for thesis titles is crucial within the academic environment of Universitas Bina Darma to prevent research duplication. The massive increase in data volume demands the implementation of mathematically efficient string matching algorithms. This study aims to compare the performance of the Brute Force and Rabin-Karp algorithms in executing exact matching on a dataset of thesis titles. A quantitative experimental approach was utilized by executing and isolating pure logic computation within a Node.js environment. Measurements focused on execution time and the number of character comparison operations across variations in keyword length, data volume, and worst-case scenarios. Experimental results demonstrate that the Brute Force algorithm consistently outperforms Rabin-Karp in terms of execution speed. In the worst-case scenario, Brute Force required only 20.0970 ms to confirm that a pattern was not found, whereas Rabin-Karp required 42.4836 ms. Based on the comprehensive empirical data, Brute Force is concluded to be the most optimal algorithm for the text characteristics of the thesis title dataset at Universitas Bina Darma, while Rabin-Karp remains advantageous in reducing the number of character comparison operations.

This study compares the performance of Decision Tree and Naive Bayes algorithms in classifying diabetes mellitus risk using the Pima Indians Diabetes Dataset (768 female patient medical records, 8 predictor attributes) without additional optimization techniques. The methodology includes data preprocessing, data splitting (80% training and 20% testing ratio), model construction, and evaluation based on accuracy, confusion matrix, and classification report, reinforced by Stratified 5-Fold Cross Validation. Hold-Out Validation results show that Decision Tree achieved an accuracy of 83.77%, while Naive Bayes achieved 76.62%. In Stratified 5-Fold Cross Validation, Decision Tree recorded an average accuracy of 85.56% (std: 0.0339), whereas Naive Bayes achieved 75.77% (std: 0.0492). Confusion matrix and classification report evaluations further confirm the superiority of Decision Tree across precision, recall, and f1-score metrics. In conclusion, the Decision Tree algorithm delivers superior and more stable performance compared to Naive Bayes in diabetes risk classification.

RJ Visual Stories faces delayed customer service due to manual handling of repetitive inquiries regarding packages, pricing, schedules, and portfolios. This study develops a conversational chatbot using Natural Language Processing (NLP) and the Multinomial Naive Bayes algorithm to provide automated customer support operating continuously (24/7). Using a Research and Development (R&D) method with a prototyping approach, the classification model was trained on 575 conversation samples across 15 intent categories utilizing TF-IDF (unigram-bigram) feature extraction. The model achieved 72.17% accuracy and a 71% weighted F1-score on the test set, with a stable average accuracy of 67.48% under 5-fold stratified cross-validation. Functional evaluation through black-box testing demonstrated a 100% success rate across 20 test scenarios. The developed system successfully automates customer inquiry handling, centralizes business information, and simplifies booking within a single chat interface, effectively eliminating administrative response delays and enhancing real-time customer engagement.

The advancement of artificial intelligence has driven a significant transformation in education, particularly through interaction-based systems designed to boost children's engagement in the learning process. These systems employ interactive approaches such as real-time responses, engaging visualizations, and content adaptation tailored to the user's abilities. However, the effectiveness of educational artificial intelligence systems depends not only on technological sophistication but also on the user experience perceived by children, the primary users. Issues arise when interface interaction designs do not fully align with children's cognitive and emotional characteristics, potentially hindering the learning process. Therefore, this study aims to evaluate the user experience of Duolingo, an AI-based language learning application, using elementary school students as a case study. The research employs a qualitative descriptive approach grounded in Child-Computer Interaction principles. Data collection involved observing task success and conducting probing interviews supported by visual instruments (such as the Smileyometer) and child-friendly evaluation guidelines. The findings are expected to provide design recommendations for educational artificial intelligence systems that are more effective, child-friendly, and capable of supporting technology-based learning.

The advancement of information technology has driven significant transformation in archive management through web-based systems designed to enhance document retrieval efficiency. These systems employ Term Frequency-Inverse Document Frequency (TF-IDF) weighting and Cosine Similarity to produce relevance-based search results and automated recommendations. However, system effectiveness depends not only on algorithmic sophistication but also on the user experience perceived by archive officers and unit leaders. This study aims to evaluate the user experience of a web-based archive management system implementing TF-IDF and Cosine Similarity at Polrestabes Palembang. The research employs a qualitative descriptive approach grounded in user experience principles across the dimensions of usability, engagement, and satisfaction. Data collection was conducted through task success observation and probing interviews supported by a Smileyometer and a simplified adaptation of the System Usability Scale (SUS). The findings revealed that although prior technical testing demonstrated high retrieval accuracy (Precision 91.30%, Recall 88.45%, and F1-Score 89.85%), this user experience evaluation identified several interface improvement needs, particularly regarding navigation clarity, terminology localization, and search result presentation.  

This study developed an image classification system for Diabetic Foot Ulcers (DFU) to distinguish between Abnormal (Ulcer) and Normal (Healthy Skin) classes using deep learning and Explainable Artificial Intelligence (XAI). An initial dataset of 2,855 images underwent resizing to 224×224 pixels, class balancing using SMOTE, partitioning with a 70:15:15 ratio, and data augmentation. SMOTE generated a balanced dataset of 4,624 images, while augmentation expanded the training set to 16,180 images. The comparative evaluation examined ResNet50 and ResNet50V2 based on accuracy, precision, recall, and F1-score. ResNet50 achieved an accuracy of 91.21%, whereas ResNet50V2 attained 98.99% with only seven misclassifications. ResNet50V2 was identified as the superior model, achieving performance metrics of approximately 99% across all evaluation criteria. Grad-CAM was implemented to provide visual interpretability through heatmaps highlighting the regions influencing the model's predictions. For ulcer classifications, Grad-CAM accurately highlighted the wound margins with output probabilities reaching 1.0. These results demonstrate that integrating ResNet50V2 with Grad-CAM produces both high diagnostic accuracy and transparent model reasoning, offering strong potential for computer-aided DFU screening, although validation on external clinical cohorts remains necessary.  

This study aims to implement the Naïve Bayes algorithm to classify food menus into "Favorite" and "Not Favorite" categories based on historical sales data at Pindang Umak KM10. The dataset consists of 275 sales records, divided into 80% (220 records) for training data and 20% (55 records) for testing data. The attributes used in the classification process include menu name, menu category, and quantity sold. The classification stages comprise data preprocessing, data splitting into training and testing sets, and the calculation of prior probability, likelihood, and posterior probability. The Naïve Bayes model was subsequently implemented in a web-based application developed using PHP and MySQL. The evaluation on 55 test records showed that 40 records were correctly classified, whereas 15 records were misclassified, achieving an accuracy of 72.73%, precision of 61.11%, recall of 95.65%, and an F1-score of 74.58%. These results indicate that the Naïve Bayes algorithm can classify food menus based on sales data patterns, thereby providing empirical support for menu management at Pindang Umak KM10. 

The Water Resources Management Agency (DPSDA) of South Sumatra Province currently determines water use permit suspension categories manually, leading to subjective and inconsistent decisions. This study aims to build a spatial analysis-based decision support system using overlay, buffer, and scoring techniques to classify suspension priority zones (high, medium, and low) and automatically generate suspension recommendations (total suspension, partial suspension, or written warning) via a web-based geographic information system. Adopting a Research and Development approach, the research utilizes water use permit data from 2020–2026 [cek data: pastikan rentang tahun izin, bandingkan dengan rujukan 2020–2023], irrigation network maps, river maps, and environmental sensitivity data. The system was developed using Laravel, PostgreSQL/PostGIS, and Leaflet.js, and was validated against 24 permit points with a minimum accuracy target of 90%. The results show that the system successfully classified suspension priority zones with an accuracy of 92%, exceeding the predetermined target. Furthermore, it generated an interactive color-classified map, automatic suspension recommendations, and downloadable reports, thereby assisting DPSDA of South Sumatra in making more objective and transparent administrative decisions.  

Coffee bean quality classification is a critical process in the agricultural industry, yet it faces significant challenges due to imbalanced data distributions that degrade standard classifiers such as Support Vector Machines (SVM). This study proposes an end-to-end machine learning framework following the CRISP-DM methodology to optimize SVM hyperparameters—specifically the penalty parameter (C) and kernel coefficient (γ)—using a continuous Genetic Algorithm (GA). A 32-dimensional visual feature extraction pipeline combining HSV color moments, RGB histograms, and grayscale texture statistics was engineered from 1,411 coffee bean images across five distinct classes (peaberry, longberry, premium, defect, and miscellaneous). To prevent data leakage, a stratified 5-fold cross-validation was implemented exclusively on 1,128 training samples to compute the cost-sensitive Macro F1-Score fitness function. The evolutionary search successfully identified optimal continuous parameters (C = 838.9254, γ = 0.0016). Experimental results on an independent holdout test set of 283 samples demonstrated that the proposed GA-SVM significantly outperformed the default baseline, elevating classification accuracy from 77.39% to 85.16% and the Macro F1-Score from 0.7742 to 0.8496. Notably, true positive detection for the commercially critical "defect" class rose from 38 to 52 instances, yielding a recall of 0.87 compared to 0.63 for the baseline. The optimized pipeline was deployed into a real-time Gradio web application featuring automated inference and Canny edge verification with latency under 150 ms, providing an accessible, interpretable decision support tool for small-scale agricultural processing facilities.

Writing scientific articles requires attention to spelling accuracy, effective communication, and the prevention of redundancy. Manually checking for spelling errors and redundant passages is often time-consuming and requires meticulous effort. This research aims to implement the Levenshtein Distance algorithm, Term Frequency-Inverse Document Frequency (TF-IDF), and Cosine Similarity within a web-based system designed for spelling validation and redundancy detection in scientific articles. The system was developed using the Web Development Life Cycle (WDLC) methodology, encompassing planning, analysis, design, implementation, testing, and maintenance. The application accepts input in the form of raw text or .docx documents. Levenshtein Distance is employed to provide spelling correction recommendations based on a reference vocabulary, while TF-IDF and Cosine Similarity are used to calculate the degree of similarity between sections of the article. Functional testing was conducted using the black-box testing method across 15 scenarios. The test results indicate that all scenarios functioned as expected. The system integrates spelling validation and similarity analysis into a single application to facilitate a structured preliminary review of scientific articles. 

The Finance and General Affairs Subdivision of the Military Court I-05 Palembang still manages office assets through manual records, physical documents, Microsoft Excel, and paper forms, resulting in fragmented asset recording, tracking, and reporting. This study aims to develop a Web-Based Office Asset Recording and Reporting System (SIMASET) for fixed and non-fixed assets. Using an Action Research approach integrated with the Web Engineering framework (communication, planning, modeling, construction, and deployment), data were gathered via observation, documentation, and interviews with three key informants between May and August 2026. The developed system accommodates asset recording, item requests, damage reporting, asset disposal, inventory recapitulation, document archiving, and user management. System evaluation using Black Box Testing across 18 test tables with 32 functional scenarios demonstrated a 100% success rate, confirming that all features operate according to the established functional requirements.

Short-term traffic condition prediction is essential for supporting responsive monitoring and proactive traffic management within intelligent transportation systems. This study aims to predict multiclass traffic conditions across four categories—normal, heavy, high, and low—for the subsequent 15-minute horizon using an eXtreme Gradient Boosting (XGBoost) model driven by temporal features. The experiment utilized a Kaggle traffic dataset comprising 2,976 chronological observations encompassing timestamp records, categorical vehicle counts, and traffic situation labels. The methodology involved data preprocessing, temporal and cyclical feature engineering, time-ordered data partitioning (2,296 training observations across the first 24 days and 671 testing observations across the final seven days), and hyperparameter optimization. Validation was performed using a three-fold TimeSeriesSplit with an observation gap to prevent temporal data leakage. Experimental results on the holdout test set demonstrated that the proposed XGBoost model achieved an overall accuracy of 75.71%, a weighted F1-score of 73.37%, a macro F1-score of 58.31%, and a balanced accuracy of 56.33%. This performance outperforms the majority baseline (56.18%) and represents an increase of 7.75 percentage points in accuracy over the persistence baseline (67.96%). Class-level analysis revealed strong predictive stability for heavy (F1-score 86.50%) and normal conditions (F1-score 82.21%), although minority classes exhibited lower sensitivity. These findings confirm that integrating XGBoost with engineered temporal features yields an effective approach for 15-minute traffic condition forecasting while rigorously maintaining chronological dependencies.

Network intrusion detection requires reliable machine learning models capable of identifying malicious traffic while maintaining valid generalization performance. However, class imbalance and cross-split feature overlap may affect the reliability of model evaluation. This study proposes a leakage-aware evaluation framework for binary network intrusion detection using the UNSW-NB15 dataset. Logistic Regression, Linear Support Vector Machine, Random Forest, and class-weighted variants of the two linear classifiers were evaluated using official and leakage-controlled test sets. A cross-split audit identified 1,302 exact overlapping feature patterns affecting 8,541 test records (10.37%), which were excluded to construct a leakage-controlled test set of 73,791 records. All classifiers exhibited performance reductions after overlap control, indicating differences between conventional and stricter generalization estimates. Random Forest achieved the best leakage-controlled performance, with 86.23% accuracy, 88.11% F1-score, 85.88% Macro F1-score, 97.68% ROC-AUC, and 97.87% average precision. It also achieved a 98.37% attack detection rate with a 1.63% false negative rate. Class weighting improved the balanced performance of both linear classifiers. These findings demonstrate the importance of jointly considering class imbalance and train-test independence when evaluating machine-learning-based intrusion detection systems.

The evolving nature of phishing attacks requires rapid and accurate detection. However, high-dimensional URL attributes can increase computational overhead during real-time processing. This study proposes a hybrid approach combining Genetic Algorithm (GA) and Random Forest (RF) for web-based phishing detection. Using 235,749 samples from the PhiUSIIL Phishing URL dataset, GA selects 20 features from the original 47 attributes, combining lexical URL characteristics with lightweight web content indicators. The dataset was partitioned using an 80:20 stratified split, yielding 188,599 training samples and 47,150 independent test samples. Evaluation on the test set yielded an accuracy of 99.66%, precision of 99.63%, recall of 99.78%, and F1-score of 99.70%, with 60 false negatives and 100 false positives. Stability testing using 10 different random seeds for data partitioning produced an average accuracy of 99.65%, with a standard deviation of 0.0241%. The model was integrated into a two-layer defense architecture using Flask and Vue.js, achieving a local server response time of less than 25 milliseconds.

Unequal household internet access remains an important dimension of the digital divide across Indonesian provinces, requiring predictive approaches that account for both socioeconomic conditions and temporal dynamics. This study aims to predict provincial household internet access and evaluate the added predictive value of previous-year connectivity. Secondary data from Statistics Indonesia covering 34 provinces during 2020–2024 were analyzed using Linear Regression, Random Forest, Gradient Boosting, and XGBoost. Two feature configurations were compared: contemporaneous socioeconomic and ICT predictors and a lag-augmented configuration incorporating previous-year internet access. Model performance was assessed using an out-of-time 2024 holdout, GroupKFold cross-validation, and rolling-origin temporal validation. Adding the lagged feature improved RMSE across all models, with XGBoost achieving the best 2024 holdout performance (RMSE = 2.7260; = 0.4003), representing a 54.49% RMSE improvement over its no-lag counterpart. Previous-year internet access was the most influential feature (55.57%), followed by adjusted expenditure per capita (21.45%) and computer ownership (18.16%). Rolling-origin validation yielded a mean of 0.5866. These findings demonstrate the value of temporal information and temporally appropriate validation for predicting regional internet access and supporting digital inclusion monitoring.

PT. Raihatul Musthofa manages hotel room assignments for umrah pilgrims, but the process is still performed manually. Staff must match pilgrims individually, which may result in mismatches among room occupants. This study implements K-Means Clustering to group umrah pilgrims based on age and region of origin and uses the resulting clusters as an initial recommendation for hotel room assignments. Data were collected through interviews and a literature review. Region of origin was transformed using label encoding, while age and encoded region were normalized using Min-Max Normalization. Clustering was performed using Euclidean Distance with K = 3 and iterated until the cluster assignments stabilized. The resulting clusters were subsequently processed using operational rules concerning gender, mahram status, and room capacity. A web-based system was developed using PHP and MySQL, providing modules for pilgrim and room management, clustering, room assignment generation, and reporting. The system was evaluated using black-box and white-box testing. Black-box testing showed that the tested functions operated as expected, while white-box testing identified 30 independent paths across the main functions summarized in the article and 56 independent paths across the entire system. The implementation provides structured, clustering-based recommendations to support hotel room assignment decisions. 

Regional disparities in digital development remain a significant challenge in Indonesia, requiring a multidimensional assessment that considers connectivity, device access, and telecommunications infrastructure. This study aims to evaluate and rank digital development across 38 Indonesian provinces using a data-driven decision support approach. Secondary data for 2024 were analyzed using an integrated Entropy-TOPSIS framework. Five initial criteria were considered, followed by correlation-based screening to reduce potential information redundancy, resulting in four final criteria: internet access, computer ownership, mobile phone ownership, and 4G/LTE share. The Entropy method assigned the highest objective weight to computer ownership (51.11%), followed by internet access (19.77%), mobile phone ownership (16.75%), and 4G/LTE share (12.37%). TOPSIS ranked DKI Jakarta first with a closeness coefficient of 0.9988, followed by DI Yogyakarta (0.7966), Kalimantan Timur (0.7721), and Kepulauan Riau (0.7691), while Papua Pegunungan recorded the lowest score (0.0096). Sensitivity analysis confirmed strong ranking stability across alternative weighting scenarios, with Spearman's correlations ranging from 0.9350 to 0.9980. The proposed framework provides a robust decision-support tool for identifying regional digital disparities and supporting more targeted digital development policies.

Public asset management at the Youth and Sports Office (Dispora) of Palembang City faces conventional administrative constraints that lead to schedule collisions (double booking), bureaucratic delays, and subjective physical condition evaluations. This study develops a web-based system, named SIPANDA, integrating the Decision Tree (C4.5) algorithm for asset physical eligibility classification and Priority Scheduling based on a Weighted Scoring Model to resolve scheduling conflicts. Developed using the Rapid Application Development (RAD) method with the Laravel framework and MySQL database, the C4.5 algorithm was trained on historical asset condition records to classify items into three status categories: fit for use, requiring maintenance, and eligible for auction, achieving a classification testing accuracy of 93.33%. Concurrently, Priority Scheduling ranks overlapping requests through weighted scoring of applicant type (60%) and activity scale (40%), employing the First In First Out (FIFO) rule as a tie-breaker. Functional black-box testing demonstrated 100% validity, the average computation time for conflict resolution reached 0.12 seconds, and User Acceptance Testing (UAT) yielded an 88.5% satisfaction score. The SIPANDA system effectively eliminates schedule collisions, enhances administrative accountability, and accelerates public asset licensing workflows.

Micro, Small, and Medium Enterprises (MSMEs) play a crucial role in the economy, yet growth is often hindered by minimal technology adoption and limited digital literacy among operators. Manual transaction and stock recording at Chicken Renggo MSME leads to operational inefficiencies and data inaccuracies. This research develops an Android-based Point of Sale (POS) and stock management application using a Human-Centered Design (HCD) approach to address this digital literacy gap. A Research and Development (R&D) methodology with Inspiration, Ideation, and Implementation phases was applied to design a cognitively-light interface using Figma, with a functional prototype built on Flutter and Laravel. The system incorporates an offline-resilient mechanism using local SQLite storage with user-triggered synchronization to preserve data transparency. Validation was conducted through Black-Box Testing confirming functional and failure-path scenarios, preliminary System Usability Scale (SUS) evaluation involving two key operational users (n = 2) resulting in an indicative score of 85 (categorized as Excellent within a micro-case study scope), supported by expert judgment triangulation from UI/UX and F&B practitioners. Results indicate that integrating HCD with offline-resilient mechanisms effectively addresses MSME needs.

Knowledge management plays a vital role in financial service organizations where operational performance relies heavily on information availability, employee expertise, and cross-divisional coordination. However, organizational knowledge is frequently undocumented, dispersed across disconnected platforms, or concentrated within specific individuals, thereby creating key person dependency and heightening organizational knowledge risk. This study investigates the factors driving key person dependency, examines how limitations in tacit knowledge transfer generate knowledge risks, and analyzes their relationship with information fragmentation at PT JACCS MPM Finance Indonesia. A qualitative case study approach was conducted through semi-structured interviews with eight purposively selected informants across the Asset, Collection, IT, Operational, and Procedure divisions, representing managerial, supervisory, and operational levels. The interview data were systematically evaluated using thematic analysis. The empirical findings reveal that tacit knowledge centralization and inadequate documentation practices not only elevate knowledge loss vulnerability during employee turnover, but also lead to operational delays, decision bottlenecks, and information fragmentation across unintegrated communication channels. To mitigate these operational disruptions, this study formulates a sustainable knowledge management strategy that integrates people, processes, and technology through centralized repository implementation, standardized documentation protocols, structured handover procedures, backup personnel assignments, and routine knowledge-sharing routines. Validated by organizational stakeholders, these strategic initiatives offer a practical framework to preserve institutional memory, ensure operational continuity, and minimize reliance on individual knowledge holders.

Learning coding and artificial intelligence requires accessible, structured, and interactive learning resources. This study aims to develop and evaluate the feasibility of a web-based digital teaching module for Grade X students at SMA Negeri 1 Palembang. This research and development study applied the 4D model comprising Define, Design, Develop, and Disseminate stages. Data on learning needs were collected through observation, interviews, documentation, and a literature review. The product was designed using Figma and implemented with HTML, CSS, and JavaScript as a responsive website. The module consists of six chapters, visual learning materials, digital flipbooks, structured navigation, and twenty practice questions per chapter. Expert validation involved 11 material experts, 10 media experts, and 8 language experts, yielding feasibility scores of 98.18%, 93.00%, and 100.00%, respectively; all fell into the highly feasible category. The module is therefore considered feasible as a supporting learning resource. Its effectiveness and direct usability for Grade X students cannot yet be concluded because valid student-response data were unavailable.