Jurnal Kecerdasan Buatan dan Teknologi Informasi

Sentiment Analysis of the Joko Widodo Diploma Controversy Using NLP and Multi-Layer Perceptron

Herdiansyah Herdiansyah — 2026-09-18

The polemic regarding the authenticity of President Joko Widodo’s diploma has become a public issue widely discussed on social media, particularly on the YouTube platform. YouTube comment sections contain various public opinions that are unstructured and written in informal language, making manual analysis difficult. Therefore, this study aims to analyze public sentiment toward the polemic of Joko Widodo’s diploma using a Natural Language Processing (NLP) approach with the Multi Layer Perceptron (MLP) algorithm. The research data were obtained from YouTube user comments related to the issue. The data were processed through text preprocessing stages, including cleaning, case folding, tokenization, stopword removal, and stemming. Furthermore, the text data were transformed into numerical representations using the Term Frequency–Inverse Document Frequency (TF-IDF) method. Sentiment classification was performed using the Multi Layer Perceptron algorithm into three sentiment classes, namely positive, negative, and neutral. Model performance evaluation was conducted using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The results show that the Multi Layer Perceptron algorithm is able to classify public sentiment with a good level of accuracy and effectively learn non-linear patterns in textual data. This study is expected to serve as a reference for the development of Indonesianlanguage sentiment analysis based on machine learning.

UI/UX Design of Simpan Kunci Password Manager: A User Centered Design Approach for Improved Usability

Ammar Arrayan — 2026-09-05

The increasing number of digital accounts forces users to manage multiple
passwords, yet many still rely on insecure storage methods, such as reusing
the same password across various platforms or manually taking notes. Furthermore,
several available password manager applications continue to face
usability and user experience issues, making them less appealing and difficult
for new users to navigate. This study aims to design the UI/UX for a Password
Manager application named “Simpan Kunci” using the User-Centered
Design (UCD) approach to ensure alignment with user requirements. The
UCD framework was applied through four structured phases: Understand
Context of Use, Specify User Requirements, Produce Design Solutions, and
Evaluate Design Against Requirements. Evaluation was conducted using the
System Usability Scale (SUS) involving 20 respondents. The research yielded
a high-fidelity UI/UX design comprising password management, master password
authentication, search functionality, and account categorization features
within a clean, intuitive interface. Based on the SUS evaluation, an average
score of 80.62 was achieved, falling into Grade B, Good adjective rating, and
Acceptable category. These results demonstrate that the proposed UI/UX
design for the Simpan Kunci application exhibits strong usability, is easy
to understand, and effectively meets user needs in managing digital account
passwords.

CashFlowMate: A Cross-Platform Mobile Application for Personal Financial Management

Nurhalisa — 2026-09-18

Personal financial recording remains a challenge for most people because it is generally done manually or solely relying on memory. This condition causes transaction data to be easily lost, is difficult to monitor, and makes it difficult for users to accurately determine income and expenditure flows. This study aims to design and build a Flutter-based personal financial recording mobile application that can record income and expenses, manage transaction categories, display financial reports in graphical form, and provide monthly installment reminders through a local notification feature. The research method used is software engineering with a Waterfall development model that includes requirements analysis, system design, implementation, testing, and maintenance. Data collection techniques were carried out through literature studies and observations of the running system, as well as system testing using black box testing and white box testing. The results of the study indicate that the application was successfully developed and all main features can function properly according to the system design, including the transaction recording process, financial report visualization, transaction history, and automatic installment reminders.

Integrated Web-Based Motorcycle Rental System with RealTime Fleet Availability Tracking

Arrizqi Hilman Rahmatulloh — 2026-01-09

Supermoto Motorcycle Rental is a motorcycle rental business that still manages vehicle data, customer records, rental transactions, and operational reports through manual processes. This condition leads to several critical issues, including difficulties in data retrieval, duplicate records, vehicle data discrepancies, delays in report generation, and the inability to monitor vehicle availability in real time. This study aims to design and implement an integrated web-based motorcycle rental information system to improve operational effectiveness and efficiency. The system was developed using the Rapid Application Development (RAD) method, encompassing requirement planning, user design, construction, and implementation phases. The application was built with the PHP programming language, the CodeIgniter 3 framework implementing the Model-View-Controller (MVC) architecture, and a MySQL database for persistent data storage. Key features include motorcycle fleet management, customer data management, rental transaction processing, payment verification, real-time vehicle availability monitoring, and automated operational reporting. System functionality was rigorously evaluated using the Black Box Testing method across 15 comprehensive test scenarios covering authentication, data management, transactions, and reporting modules. The testing results demonstrated a 100% success rate across all evaluated scenarios, confirming that all system features function correctly and fully meet the specified user requirements. The developed system successfully streamlines data management, enhances data accuracy, accelerates administrative workflows, and provides real-time operational monitoring capabilities. Ultimately, this digital transformation equips Supermoto Motorcycle Rental with a reliable, secure, and scalable platform to support data-driven decision-making and elevate overall customer service quality and long-term business competitiveness.

Optimization of the Decision Tree Algorithm Using SMOTE for the Classification of Diabetes Mellitus Types Based on Clinical Medical Record Data

Ia Riham Nurrahma — 2026-09-01

Diabetes mellitus is a chronic metabolic disorder characterized by elevated blood glucose levels. Accurate early classification of diabetes types is crucial for determining appropriate clinical interventions. However, clinical datasets often suffer from severe class imbalance, leading to model bias toward the majority class. This study proposes the optimization of the Decision Tree classification algorithm by integrating the Synthetic Minority Over-sampling Technique (SMOTE) using real-world clinical medical record data from the Singaparna Community Health Center. The dataset comprises 225 patient records with 38 health attributes. Preprocessing and attribute transformation techniques were applied, followed by stratified random sampling (80:20 split). To resolve the extreme class imbalance (1 Type 1 DM vs 224 Type 2 DM samples), SMOTE was applied exclusively to the training set, balancing the dataset to 358 instances (179 samples per class). The Decision Tree model was trained and evaluated using validation curves to prevent overfitting. The experimental results demonstrate that the SMOTE-optimized Decision Tree model with an optimal max depth of 3 achieved robust classification performance, yielding an accuracy of 86.67%, precision of 88.54%, recall of 84.12%, F1-score of 85.20%, and an ROC-AUC score of 0.9333. Furthermore, the decision tree visualization revealed transparent, age-based rule splits matching clinical domain knowledge. This study confirms that SMOTE effectively mitigates class imbalance in clinical decision-tree modeling, providing a reliable and interpretable decision-support tool for diabetes screening in primary healthcare settings.

Scrum Board Application Development Using MERN via Scrum Method and AHP Algorithm for Software Project Management

Hafiyan Rizqi Sanjaya — 2026-09-01

Agile Scrum is widely adopted in software engineering, yet teams frequently face challenges such as unstructured backlog management, limited sprint progress visibility, and highly subjective priority decisions. This study addresses these critical issues by developing an Android-based project management application that integrates Scrum Board functionality with the Analytical Hierarchy Process (AHP) for objective and measurable backlog prioritization. Built on the MERN Stack architecture (MongoDB, Express.js, React Native, Node.js), the system was developed iteratively through four one-week sprint cycles. The integrated AHP module evaluates backlog items using three weighted criteria: Urgency (0.65), Complexity (0.23), and Work Value (0.12). System correctness and reliability were rigorously validated using Black Box Testing methodologies. The results demonstrate that the AHP module yields a highly consistent priority ranking (Consistency Ratio = 0.017), successfully identifying the Login Feature as the highest priority with a score of 4.42. Furthermore, functional testing confirmed 100% compliance across all core scenarios, including user authentication, sprint creation, task assignment, and Scrum Board status updates. This research successfully unifies Agile Scrum management and AHP decision support within a single, cohesive mobile application. By embedding objective decision-making directly into the daily project workflow, the system effectively eliminates reliance on external prioritization tools, bridging the current project management tool gap. Ultimately, it provides a scalable, replicable architecture suitable for future enhancements, such as real-time collaboration, push notifications, and multi-project portfolio management.

Software Development Safety Monitoring Interface for Elevator Systems Using Bitwise Decoding and Finite State Machines

Ilham maulana — 2026-09-01

This study develops an Elevator Safety Monitoring Interface that integrates a Bitwise Decoding method and a Finite State Machine (FSM) algorithm to monitor elevator operational conditions in real-time. Modern elevator systems rely on the integration of electromechanical components and embedded systems; however, condition monitoring is often performed manually, leading to limited real-time information, increased occupational health and safety (OHS) risks, and the absence of structured historical fault records. This study aims to design a VB.NET-based monitoring system capable of displaying elevator status in real-time, implementing a Bitwise Decoding method for efficient processing of binary data from a microcontroller, and applying the FSM algorithm to model elevator operational conditions deterministically. The Waterfall development model is employed, encompassing planning, analysis, design, implementation, and testing phases. Hardware used includes Arduino/ESP32 microcontroller, relay circuits as digital signal inputs, RS485 module, and an RS-485 to USB converter. The monitoring software was developed using VB.NET with MySQL for activity log storage. The FSM is formally defined as a 5-tuple M = (S, Σ, δ, s 0 , F), with states including IDLE, MOVING_UP, MOVING_DOWN, DOOR_OPEN, FIRE_MODE, INSPECTION, EARTHQUAKE_MODE, ERROR, and OFFLINE. Results demonstrate that the system successfully displays elevator operational status in real-time and effectively extracts elevator condition information from binary data with high efficiency and low latency. The developed monitoring system is expected to improve technician work efficiency and support elevator operational safety.

Implementation of An IoT-Based Temperature and Humidity Monitoring System with Data Logging Features in the Calibration Laboratory

Ryan Franki Risdianto — 2026-09-01

ISO/IEC 17025:2017 accredited calibration laboratory, holds an absolute obligation to monitor, control, and record environmental conditions (temperature and humidity) to maintain the validity of measurement results and support budget uncertainty calculations. However, the current monitoring system is still conventional and manual (three times a day), making it vulnerable to human errors, data duplication, and an inability to record fluctuations between intervals. This study aims to design and build an automatic temperature and humidity monitoring system based on the Internet of Things (IoT) featuring a dual-logging system to ensure continuous technical records. The system design utilizes an ESP32 microcontroller as the central processing unit, a DHT11 sensor to detect ambient temperature and relative humidity, a Micro SD Card module as a local backup media (offline backup), and the ThingSpeak platform as a cloud database for remote monitoring. The sensor data is saved to the SD Card in CSV format every 5 minutes, while data transmission to ThingSpeak is scheduled every 1 minute. System testing involved functionality tests (Black Box Testing), local storage capability tests, connectivity tests, data validity tests, and stability/durability tests. The results indicate that the IoT-based monitoring device was successfully functionality test success rate. The implementation of the dual-logging feature effectively preserves the integrity and continuity of records; ThingSpeak simplifies realtime data visualization, while the Micro SD Card operates as a reliable local failsafe during network fluctuations. This system minimizes human error, increases operational oversight efficiency, and supports regulatory compliance with Clauses 6.3 and 7.5 of ISO/IEC 17025:2017.

Optimization of Fuzzy Tsukamoto for Waste Paper Quality Unfitness Weighting

Fairuz Hadi — 2026-09-01

Waste paper has economic value because it can be reused as raw material in the recycling industry. However, the assessment of waste paper quality in purchasing transactions is often still performed manually and may lead to subjective decisions, especially when several variables must be considered simultaneously. This study aims to apply the Fuzzy Tsukamoto method to determine the unfitness weight of waste paper quality as a supporting basis for determining the payable weight in waste paper purchasing transactions. The research was conducted at PT Gusti Putra using an applied research approach with system development. Data were collected through observation, interviews, and literature study. The variables used in the fuzzy calculation consist of vehicle type, gross weight, net weight, and paper quality as input variables, while the output variable is the percentage of waste paper quality unfitness weight. The calculation process includes fuzzification, fuzzy rule formation, inference using the AND operator, and defuzzification using the weighted average method. A case study was conducted using vehicle type K2, gross weight of 1600 kg, net weight of 1150 kg, and paper quality value of 6. The result shows an unfitness weight value of 13.04%, which belongs to the medium category. This result indicates that the Fuzzy Tsukamoto method can provide a more systematic, objective, and measurable assessment of waste paper quality and can support the payable weight calculation process in waste paper purchasing transactions.

Stress Monitor Design and Construction Based on Arduino Uno

Bayu Wahyudi — 2026-09-01

Monitoring physiological conditions such as heart rate during high physical activities, including stress test monitoring, is highly crucial to prevent cardiac overload. However, conventional stress test systems rely on expensive equipment, limiting their accessibility in small healthcare facilities. This study aims to design and construct an Arduino Uno-based stress test monitor utilizing an AD8232 ECG sensor integrated into a modified treadmill system. The system employs an Arduino Uno microcontroller as the primary data processor to analyze the bioelectric heart signals captured by the AD8232 sensor, displaying the real-time results on a TFT screen. Device performance was evaluated through voltage measurements, functional testing, speed calibration, and comparative testing across three speed levels: low mode, medium mode, and high mode. The results showed that the power distribution system operated stably with a maximum voltage error of 1.81%. Speed calibration using a tachometer produced three distinct intensity levels of 16 rpm, 24.6 rpm, and 33 rpm. Comparative testing against a standard pulse oximeter showed average heart rate readings of 98 BPM in low mode, 125 BPM in medium mode, and 155 BPM in high mode, with a percentage error of 0% and a correction value of 0 BPM across all modes. These results indicate excellent device accuracy, with deviations remaining well within acceptable tolerance limits. Consequently, the designed system functions as intended and is declared eligible for operational use to support user physical performance monitoring.

Fraud Detection in Subsidized LPG Gas Distribution at the Base Level as a Monitoring System Using the Mamdani Fuzzy Logic Algorithm

M.Syahrul — 2026-09-01

he distribution of subsidized Liquefied Petroleum Gas (LPG) at pangkalan levels is still prone to fraud, such as discrepancies between stock and distribution volumes, sales above the maximum retail price (HET), and undelivered distributions, all of which are still mostly supervised manually. These conditions make it difficult to quickly and objectively detect the presence of a fraudulent distribution pattern. This research develops a Mamdani Fuzzy Logic-based fraud detection engine that functions as a real-time monitoring system for subsidized LPG gas distribution. The system was developed using the Waterfall method, which includes requirements analysis, design, implementation, and testing, and was built with Laravel, React.js, and Inertia.js as a Single Page Application (SPA) architecture. Fraud-indication analysis involves three fuzzy input variables — remaining stock, daily distribution volume, and the Purchase Order (PO) submitted to the agent. They are processed through fuzzification, a 27-rule rule base, MIN-MAX inference, and Centroid of Area defuzzification. The output is three categories: safe (aman), alert (waspada), and irregular/indicative of fraud (menyimpang). Real transaction data from Pangkalan Gas Asmalia for May 2026 shows that a daily distribution of 18 cylinders against a daily quota of 46 cylinders resulted in a crisp output of 16.20, which falls under the Safe category and is in line with actual field conditions. This system would enable the pangkalan operators to identify early fraud signals, have more structured and objective decisionmaking, and enhance the accountability of subsidized LPG distribution.

Implementation of the Density-Based Spatial Clustering (DBSCAN) Method to Identify Beauty Clinic Customer Purchasing Patterns

Taufik Rifki Al Hafiz Lubis — 2026-09-01

Beauty clinics routinely accumulate detailed transaction records, yet these data are often used only for administrative reporting and are not transformed into actionable customer insights. This study implements Density-Based Spatial Clustering of Applications with Noise (DBSCAN) in a web-based system to identify customer purchasing patterns at Raen Aesthetic. Transaction data from January to December 2025 were cleaned, aggregated into recency, frequency, and monetary features, and normalized using a robust scaler. DBSCAN was executed with epsilon = 0.8000 and minimum samples = 5, while cluster quality was assessed using the Silhouette Score and Davies-Bouldin Index. The implementation produced nine customer clusters and 148 noise observations. The Silhouette Score of 0.3551 indicated a reasonably structured separation, whereas the Davies-Bouldin Index was 1.2569. Cluster 1 was the largest segment, comprising 355 customers with low purchase frequency and relatively small monetary value. Cluster 5 represented a high-value segment with an average monetary value of approximately IDR 10,302,000, while the noise group displayed heterogeneous but economically significant transaction behavior. Treatment transactions dominated most clusters, whereas several segments showed stronger preferences for packages and deposits. Black-box testing confirmed that all major functions operated according to the specified scenarios. The results demonstrate that DBSCAN can provide interpretable customer segmentation and support evidence-based service and customer-management decisions in beauty clinics.

SOA-Based Digitalisation of the Jakarta Metropolitan Police’s ICT Services for the Integrity of the Indonesian National Police’s Internal Services

Oktavio Jati Moris Pereira — 2026-09-01

Internal Information and Communication Technology (ICT) services within the Jakarta Metropolitan Police have traditionally relied on manual, fragmented processes, resulting in data duplication, delayed reporting, and weak accountability in IT asset management. This study aims to digitize these internal technical services and enhance organizational integrity by developing a centralized web-based application using a Service-Oriented Architecture (SOA). Employing a descriptive qualitative approach, data were gathered through observation, interviews, and documentation to map and restructure existing business workflows. The resulting system, Lantik PMJ, was built using CodeIgniter 4, MySQL, and Bootstrap 4 to integrate modular services for IT equipment loans, network installations, device repairs, and comprehensive asset lifecycle tracking. Black-box testing was conducted to validate system functionality across tiered, role-based access levels. The results demonstrate that the SOA-based application successfully replaces informal communication channels with a structured workflow, providing real-time service tracking, automated PDF reporting, and proactive return-delay monitoring for IT assets. All tested functional scenarios passed successfully, confirming the system’s reliability and readiness for operational deployment. Ultimately, this digital transformation significantly improves operational efficiency, transparency, and internal accountability, serving as a foundational model for the broader modernization of the Indonesian National Police’s internal support infrastructure.

Transformation of Basketball Academy Athlete Performance Evaluation Using Information Technology and the Weighted Product Method

Pahrizal — 2026-09-01

Performance evaluation of basketball athletes at academies in Indonesia still largely relies on manual, subjective, and unintegrated assessment processes, leading to inconsistencies between evaluation periods. This study aims to design and develop a web-based information system for athlete performance evaluation at Bendella Basketball Court, Bekasi, using the Weighted Product (WP) algorithm. The system was developed following the Waterfall Software Development Life Cycle (SDLC) methodology, covering requirements analysis, system design, implementation with the Laravel framework, and testing. Eight evaluation criteria were established: shooting (20%), dribbling (15%), defense (15%), passing (10%), speed (10%), stamina (10%), discipline (10%), and teamwork (10%). The WP algorithm computes vector S and preference value V for each athlete, followed by threshold-based categorization into Junior Advance and Junior Basic categories. Testing on five sample athletes demonstrates that the system correctly ranks and categorizes athletes objectively. The Weighted Product method provides greater stability compared to Simple Additive Weighting (SAW) with a sensitivity change of only 0.01% versus 0.38%. This system supports coaches in making more accurate, consistent, and data-driven athlete evaluation decisions.

Optimization of Spare Parts Stock Control Using Safety Stock and Reorder Point Approach with Laravel

Dimas Sanjaya — 2026-09-01

Inventory management is a critical aspect of ensuring business operational continuity, particularly in maintaining spare parts availability for automotive workshops. Bengkel Mo Gerzz, a Vespa Matic specialist workshop, handles 1 to 10 vehicles daily with over 100 types of spare parts managed manually through simple records and verbal communication, leading to real-time stock monitoring limitations and reorder inaccuracies. This study develops a web-based inventory management system built with the Laravel framework, integrating the Safety Stock (SS) and Reorder Point (ROP) methods based on lead time to optimize spare parts stock control. The system was developed using the Waterfall methodology through seven sequential phases: planning, analysis, design, development, testing, implementation, and maintenance. The proposed SS formula calculates buffer stock as SS = (Dmax × Lmax) − (Davg × Lavg), while ROP is determined by ROP = (Davg × L) + SS. The system was evaluated using Black Box Testing across six modules, achieving a 100% pass rate across 24 test scenarios. Calculation results for 21 spare part types showed that Oli Mesin Ipone Scooter 10W-40 Premium recorded the highest SS value of 14.97 and ROP of 18, indicating the greatest demand variability. The system provides automated restock notifications when stock reaches the ROP threshold, enabling more structured and data-driven procurement decisions at Bengkel Mo Gerzz.

Application of K-Means Clustering Algorithm in Edam Burger Sales Information System for Inventory Control Optimization

Muhammad Rofiq Ubaidillah — 2026-09-01

Manual sales and inventory management in small culinary enterprises often leads to data inaccuracies, stock mismanagement, and underutilized transactional data. This study aims to design a web-based sales information system integrated with the K-Means clustering algorithm to optimize inventory control at Edam Burger & Frozen Foods. Utilizing the Waterfall methodology, the system was developed using the Laravel framework and MySQL. The analytical engine processed five months of transactional data across fifteen products, applying Min-Max Normalization to equalize the scales of sales volume, revenue, and transaction frequency. The K-Means algorithm successfully segmented the product catalog into three distinct categories based on performance: one high-selling core product (6.7%), three medium-selling secondary items (20%), and eleven low-selling complementary products (73.3%). Black Box Testing confirmed a 100% functional success rate across all system modules. The primary novelty of this research lies in seamlessly embedding the K-Means engine directly into the operational dashboard, overcoming the common barrier of offline, standalone data mining. This integration enables real-time, data-driven procurement strategies, providing actionable recommendations: prioritizing continuous stock availability for high-demand items, scheduling regular restocking for medium items, and minimizing capital tied up in low-moving inventory to reduce food waste. Ultimately, this integrated approach empowers small business owners to transition from intuition-based management to systematic, algorithm-driven inventory optimization. This study successfully bridges the gap between routine transactions and strategic analytics.

Transforming E-Commerce UX: A User-Centered Design Approach to Enhance Usability and Web Performance

MY Aura Puti Agdira — 2026-09-01

E-commerce platforms require optimal user interface (UI) and user experience
(UX) alongside robust technical performance to retain users, particularly in
specialized markets like mother-and-baby products. Initially, the Boonaboo
e-commerce website exhibited critical deficiencies, achieving a poor System
Usability Scale (SUS) score of 47.75 and a failing mobile Google Lighthouse
performance score of 41. To address these dual challenges, this study implemented
a comprehensive UI/UX redesign using the User-Centered Design
(UCD) methodology. A mixed-methods, pretest-posttest evaluation was conducted
involving 20 respondents, utilizing a novel dual-metric approach that
combined subjective usability measurement (SUS) with objective technical
auditing (Core Web Vitals via Google Lighthouse). Following the UCD
intervention—which encompassed information architecture restructuring, navigation
simplification, and asset optimization—the platform demonstrated
remarkable enhancements. Mobile performance surged by 104.8% (from 41 to
84), while desktop performance reached near-perfection at 99. Concurrently,
the mean SUS score escalated to 87.75, shifting the usability grade from Not
Acceptable (Grade F) to Excellent (Grade A). These findings empirically prove
that user-centered interface improvements inherently correlate with enhanced
machine-measured web performance. This study contributes a validated dualmetric
framework for evaluating e-commerce redesigns, offering actionable
insights for optimizing digital retail environments where cognitive ease, speed,
and user trust are paramount.

Development of a Web-Based Health Information System Integrated with Artificial Intelligence Using Agile Methodology at Dasan Agung Primary Health Center

Muhamad Masjun Efendi — 2026-09-01

The recording of health data at the Dasan Agung Community Health Center
(Puskesmas) in Mataram City is still carried out manually, giving rise to
various problems such as data entry errors, data loss, delays in data retrieval,
and inaccuracies in health reports. This study aims to develop a Web-Based
Health Data Recording Information System using the Agile method to enhance
the effectiveness of health data management at the community health center.
The method employed encompasses the stages of requirements gathering, analysis,
design, coding, testing, deployment, and feedback. Data collection was
conducted through observation, interviews, and documentation studies. The
system was built using the PHP programming language with the CodeIgniter
3.1.13 framework and a MySQL database. The findings indicate that the developed
system is able to facilitate a more effective, faster, and integrated health
data recording process. Available features include patient data management,
personnel data management, patient registration, drug data management,
laboratory management, medical records, patient prescriptions, and digital
health reports. With the implementation of this system, health administration
processes become more optimal and support the digital transformation of
health services at the community health center. The outputs of this research
are a web-based information system and the publication of a scientific article
in a national journal.

Web-Based Stock Overstock Warning System for Spare Parts Inventory Using Linear Regression

Rizky Fadillah Putra Pratama — 2026-09-01

Inventory management in the automotive spare parts industry faces a critical
challenge in managing overstock conditions, which lead to increased storage
costs and capital freezing. PT. Dipo Internasional Pahala Otomotif currently
manages spare parts inventory manually, without any predictive alert system
capable of detecting potential overstock based on sales data. This study develops
a web-based overstock warning system using the Simple Linear Regression
algorithm implemented in the Laravel framework to predict spare parts stock
requirements and automatically trigger overstock alerts. The system was built
following the Waterfall development methodology through seven sequential
phases: planning, analysis, design, coding, testing, implementation, and maintenance.
The linear regression model uses time period as the independent
variable (X) and stock quantity as the dependent variable (Y ), forming the
prediction equation Y = a + bX. Based on a simulation with n = 4 periods,
the resulting equation Y = 7 + 2.7X predicted a stock of 20.5 units in period
5, which exceeded the defined overstock threshold. System evaluation using
Black Box Testing confirmed that all functional modules operated correctly.
The system successfully provides automated overstock detection and real-time
alert notifications, enabling more accurate and data-driven inventory decisions
at PT. Dipo Internasional Pahala Otomotif.

Interpreting Text-Enriched Dual-Head Multitask Learning for Indonesian Hateful Meme Detection Using Explainable AI

Selamet Riadi — 2026-09-01

Internet memes in Indonesia are frequently weaponized to disseminate implicit
hate speech through sarcasm and cultural nuances. Automatically detecting
such content is computationally challenging, and existing deep learning
frameworks predominantly operate as opaque black boxes, lacking decision
transparency. This study implements and optimizes a text-enriched dual-head
multitask learning architecture utilizing IndoBERTweet to concurrently classify
hatefulness and appropriateness within the INDOMEME dataset. Rather
than processing raw image pixels, we employ a text-enrichment strategy where
visual semantics are transcribed into textual descriptors via Optical Character
Recognition and vision-language captioning. To bridge the interpretability gap,
we deploy Local Interpretable Model-agnostic Explanations (LIME) to decode
the internal feature attributions of the architecture. Furthermore, advanced
training optimizations, encompassing cosine annealing, gradient accumulation,
class-weighted loss, and dynamic threshold calibration, were engineered to
enhance model generalization. Experimental evaluations demonstrate that
the optimized model achieves a Macro-F1 score of 0.812 for hatefulness and
0.820 for appropriateness, surpassing the established baseline. Crucially, the
LIME analysis unveils a pivotal finding: despite sharing an identical textual
backbone, the hate-specific head predominantly focuses on lexicons carrying
social agitation, whereas the appropriateness head prioritizes general norm
violations. These empirical findings substantiate that multitask learning enriches
semantic representation quality, offering a transparent framework for
trustworthy content moderation.

A Web-Based Credit Financing Eligibility Determination System for Multifinance Companies Using the Fuzzy Tsukamoto Method

Al Ihsan Fauzi Ardilla — 2026-09-01

The credit eligibility analysis process in a financing company is a crucial stage in
minimizing the risk of non-performing loans. However, the assessment process
that still relies on the subjective interpretation of credit analysts potentially
leads to inconsistencies in decision-making. This issue also occurs at PT
Mandiri Utama Finance Bekasi Branch, which has shown a declining trend in
financing performance based on the indicators of End Net Receivables (ENR),
Total Over Due (TOD), and Net Credit Loss (NCL). This study aims to design
and develop a web-based decision support system that can assist in determining
credit financing eligibility by applying the Fuzzy Tsukamoto method based on
the 5C principles (Character, Capacity, Capital, Collateral, and Condition).
The research methods employed include observation, interviews, and literature
studies. System development was carried out using the Waterfall model,
comprising requirement analysis, system design, implementation, testing,
deployment, and maintenance phases. The Fuzzy Tsukamoto method was
utilized to process data containing uncertainty through fuzzification, IF-THEN
rule-based inference, and defuzzification to generate credit eligibility scores.
The expected result of this study is the creation of a web-based decision
support system capable of providing credit eligibility recommendations that
are more objective, consistent, and structured. Consequently, the developed
system can assist credit analysts in the decision-making process and support
the company’s efforts to minimize the risk of problematic financing.

Design and Development of a Web-Based Disaster Health Information System Prototype for Medical Record and Health Information Students

Mochammad Arief Darmawan — 2026-09-01

Disaster-response documentation requires rapid linkage of victim, healthservice,
logistics, and reporting data, while Medical Record and Health Information
students have limited access to disaster-specific learning environments.
This study designed a web-based Disaster Health Information System prototype
as an educational simulation for disaster health information management.
Requirements were synthesized from disaster-documentation needs, policy and
standards, prior literature, and intended student learning workflows; the study
did not report interview-, focus-group-, or questionnaire-based requirements
elicitation and did not conduct respondent usability testing. An iterative
prototype cycle covered quick design, DFD/ERD modeling, interface development,
and scenario-based design verification with synthetic display data. The
prototype established a role-based learning workflow that connects disaster
events with victim records, health services, urgent logistics, dashboard summaries,
and crisis reporting. Design verification confirmed traceability among
major processes and seven core data entities, while interface evidence was
strongest for login and dashboard functions; other modules were verified at the
design-traceability level. The educational novelty is the structured simulation
workflow linking patient-level documentation with event-level logistics and
reporting, rather than the individual use of dashboards, DFDs, ERDs, or rolebased
access. The prototype demonstrates a coherent learning-oriented design
but does not establish usability, performance, interoperability, or operational
effectiveness.

K-Means Clustering for Drug Inventory Analysis at Anugrah Pharmacy Bekasi

Alya Prciscilla putri — 2026-09-01

Improper drug inventory management can cause stockouts, overstocking, and
inefficient procurement decisions, especially in small-scale retail pharmacies
that still rely on manual estimation. This study analyzes drug sales patterns
and groups drug inventory at Anugrah Pharmacy Bekasi using the K-Means
clustering algorithm within the Cross Industry Standard Process for Data
Mining (CRISP-DM) framework. The dataset consisted of 5,312 sales transactions
from January to June 2025. The transaction records were aggregated
into 731 drug items using three variables: transaction frequency, sales volume,
and transaction value. Data preparation included aggregation, missing-value
checking, duplicate checking, transformation, and Min-Max normalization.
The optimal number of clusters was determined using the Elbow Method,
which indicated three clusters (k = 3). The K-Means results grouped the 731
drug items into 28 Fast Moving items (3.83%), 129 Medium Moving items
(17.65%), and 574 Slow Moving items (78.52%). The centroid analysis shows
that each cluster has distinct sales-movement characteristics. The results can
support inventory decision-making by helping the pharmacy prioritize replenishment
for fast-moving drugs, maintain controlled stock for medium-moving
drugs, and limit excessive procurement for slow-moving drugs. This study
demonstrates that CRISP-DM and K-Means clustering can provide practical
information for data-driven drug inventory management in a retail pharmacy
context.

Cosmetic Product Segmentation Analysis Using K-Means Clustering at PT Mandom Bekasi

Mona Dewintha Agustine — 2026-09-01

PT Mandom Indonesia Tbk manages a wide range of cosmetic products with varying sales levels and inventory turnover rates, creating challenges in inventory management and marketing strategy formulation. This study aims to segment cosmetic products based on sales patterns and inventory turnover using the K-Means Clustering algorithm within a Knowledge Discovery in Databases (KDD) framework. The research stages include data selection, preprocessing, transformation, clustering, and evaluation. The dataset consists of 436 cosmetic products with attributes including sell in, sell out, stock, and expiration date, sourced from PT Mandom's internal sales report for the year 2025. Feature engineering produced two derived variables, the sell out to sell in ratio and the remaining days until expiration, which were normalized using Min-Max Scaling. The optimal number of clusters, determined using the Elbow Method and validated with the Silhouette Score, was three. The K-Means algorithm successfully grouped the products into three segments: Fast Moving (75 products, 17.2%), Medium Moving (299 products, 68.6%), and Slow Moving (62 products, 14.2%). The Fast Moving cluster exhibited the highest sell in, sell out, and sell-through ratio values, while the Slow Moving cluster showed the lowest ratio, indicating a higher risk of stock accumulation. These segmentation results can serve as a data-driven basis for inventory management, distribution planning, and marketing strategy decisions at PT Mandom.

Implementation of Single Moving Average Algorithm for Maintenance Material Prediction inWarehouse Supply Chain System

Steven Aditya Pratama — 2026-09-01

Material inventory management at a telecommunication industry warehouse frequently experiences imbalances due to unpredictable monthly demand fluctuations, triggering the risk of material shortages or excessive stockpiling. This study aims to implement predictive computation into the supply chain information system to forecast the volume of material requirements for maintenance work. The quantitative forecasting method used is the Single Moving Average (SMA) algorithm, which extracts actual historical expenditure data. The accuracy level of the system's projection results is evaluated mathematically using the Mean Absolute Deviation (MAD) instrument. Furthermore, the system was developed using the Software Development Life Cycle (SDLC) Waterfall model and the CodeIgniter framework. Testing was conducted by comparing the moving average parameters for a three-month (=3) and a six-month (=6) period. The system's computation results on operational data show that the algorithm can dynamically calculate projections with MAD error values that vary depending on the fluctuation of the material type, such as ODP forecasting (=3) recording a MAD deviation of 1.06, and Iron Poles (=6) with a MAD of 0.83. This study proves that integrating the SMA algorithm into the warehouse database can serve as a reliable reference parameter for management in determining measurable material procurement volumes.

Optical Marketplace and POS Integration Based on Content- Based Filtering

Dystian En Yusgiantoro — 2026-09-01

Manual transaction management in optical businesses often results in stock data discrepancies, delays in report preparation, and limited marketing reach. This study aims to develop an optical marketplace integrated with a Point of Sales (PoS) system to synchronize transaction and stock data in real-time. In addition, a Content-Based Filtering algorithm is implemented with binary product attribute weighting and customer preference profiles based on transaction frequency, combined with multi-channel Cosine Similarity calculations to generate product recommendations based on product characteristics and customer purchase history. The study was conducted through the stages of needs analysis, system design, implementation, and functional testing using the Black-box Testing method. The results show that the integration of the marketplace and PoS successfully maintains the consistency of transaction and stock data, while the Content-Based Filtering algorithm is able to provide product recommendations that match customer preferences. This study shows that the integration of the marketplace, Point of Sales, and recommendation systems can support the digitalization of optical businesses through more efficient data management and an improved customer shopping experience.