Jurnal CoSciTech (Computer Science and Information Technology)

Model Deteksi Teks Generatif menggunakan Stacking Ensemble Berbasis DistilBERT dan Few-Shot Adaptation

2026-08-16T06:43:59+07:00

The rapid development of Generative Artificial Intelligence (AI) technologies such as ChatGPT, Claude, and Gemini has brought new challenges to academic integrity in higher education. Detection of AI-generated written works has become crucial to ensure the originality and intellectual honesty of students. However, conventional AI detectors are often static and have low accuracy against dynamic variations in writing styles. This research aims to develop a more adaptive and accurate generative text detection system by applying the Stacking Ensemble Learning method. This system integrates four basic classification models (Logistic Regression, SVM, MLP, and Random Forest) enhanced with high-level semantic features using DistilBERT Embeddings. In addition, this system utilizes the Google Gemini API as a feature augmentation instrument to obtain real-time qualitative risk assessments. The main innovation in this research lies in the application of the Few-Shot Adaptation technique. This feature allows lecturers or users in higher education to instantly recalibrate the Meta-Classifier by simply uploading a few examples of original documents (human-written) and AI-generated documents. The system was developed web-based using the Gradio framework, which can process various academic document formats such as .pdf, .docx, and .txt. The results of this research are expected to provide a tangible contribution to higher education institutions in the form of a digital instrument capable of precisely validating the authenticity of academic documents. In addition to providing a probability score, the system also includes a Text Highlighting feature to mark suspicious text sections, thus simplifying the evaluation process by educators.

Keywords: AI Detection, Stacking Ensemble, DistilBERT, Few-Shot Adaptation, LLM.

Penerapan Model ARIMAX untuk Prediksi Penjualan Toko Kiranashoop15 di Shopee

2026-08-19T15:23:29+07:00

Transactions on the Shopee marketplace generate large volumes of records that can be turned into a basis for sales forecasting. This study applies an Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) approach to forecast the monthly total sales of the Kiranashoop15 store while identifying the exogenous factors that most influence the model. Data were drawn from the Shopee Seller Centre income report for January 2020 to May 2025, originally 11,142 order-level records, which were cleaned and aggregated into 65 monthly observations. The series was split chronologically into 52 training months and 13 testing months, with standardized exogenous variables covering product price, discounts, vouchers, cashback, shipping cost, and marketplace fees. Model parameters were searched through a grid search minimizing the Akaike Information Criterion (AIC). The selected model, ARIMAX(2,1,4), produced an AIC of 1203.78. On the test set the model achieved an MAE of Rp51,825, RMSE of Rp62,484, and MAPE of 0.461%, while the Ljung-Box test (p-value 1.00) confirmed white-noise residuals. Product Original Price contributed the largest relative share (68.87%), followed by Campaign Cost (22.13%). The six-month forecast (June-November 2025) indicates total sales ranging from Rp14.20 million to Rp15.22 million per month with a stable tendency. These findings confirm that ARIMAX is a suitable tool for marketplace sales forecasting.

Implementation of a Vision Transformer for Web-Based Acne Type Classification Using Facial Images

2026-08-25T12:00:18+07:00

Acne Vulgaris is a widespread skin condition that can affect not only physical skin health but also a person's self-confidence. Because manually distinguishing acne types still demands specialized expertise, an artificial-intelligence-based approach is needed to support this classification task. This research applies a Vision Transformer (ViT-B/16) architecture to categorize acne lesions from facial images into four groups: normal, papule, pustule, and nodule. A total of 4,000 images were used as the dataset and processed through a transfer-learning strategy initialized with pre-trained ImageNet weights. The model was trained across 20 epochs with the Adam optimizer and a learning rate of 0.001. Testing showed that the model reached an accuracy of 86%. The resulting model was then embedded into a web-based application to streamline the acne identification workflow. These outcomes confirm that Vision Transformer can classify acne types reliably and holds promise as an automated early-screening tool for facial skin conditions.

Access Control Box System with Wi-Fi Manager on ESP32 and Fingerprint Authentication Based on Android Application

2026-08-15T16:29:23+07:00

A storage box is an essential container for securing valuable items and offers the flexibility to be placed in various locations. This research aims to implement the AC-Box (Access Control Box) system based on the Internet of Things (IoT), focusing on ease of network configuration and access security. The system utilizes an ESP32 integrated with a Wi-Fi Manager feature, allowing the device to connect to new networks dynamically without the need for reprogramming (hardcoding). For the security mechanism, fingerprint authentication is employed via the smartphone's built-in biometric sensor, which is connected to an Android application as the user interface. Based on testing results across two network types—a ZTE Wi-Fi access point and a smartphone hotspot—the system performs adequately in responding to every unlock command. The average delay on the ZTE Wi-Fi access point was 384.2 ms, categorized as "Medium," while the smartphone hotspot recorded 566.7 ms, categorized as "Poor" according to TIPHON standards. The initial connection time via Wi-Fi Manager ranged from 806.1 ms to 3,151.4 ms. Fingerprint authentication security testing successfully recognized 5 registered fingerprints and immediately activated the solenoid lock. Conversely, 25 unregistered fingerprints were rejected by the system, resulting in no commands being sent by the application and the box remaining locked. These results prove that the system can accurately validate access and prevent unauthorized entry. In conclusion, the AC-Box serves as a practical, secure, and efficient access system solution for personal storage needs.

Prediction of rubber commodity prices using support vector regression (SVR)

2026-08-15T16:41:32+07:00

Natural rubber prices constitute a time-series dataset characterized by nonlinear and volatile patterns, making the prediction process challenging. This study aims to evaluate the performance of Support Vector Regression (SVR) with a Radial Basis Function (RBF) kernel and to examine the effect of hyperparameter optimization using Grid Search on natural rubber price prediction. The dataset consists of monthly global rubber prices from January 2000 to December 2020 obtained from IndexMundi, comprising a total of 252 observations. The study was conducted using two scenarios: SVR without hyperparameter optimization and SVR with hyperparameter optimization using Grid Search. Model performance was evaluated using Root Mean Squared Error (RMSE) on the testing data covering the period from January 2018 to December 2020. The results indicate that hyperparameter optimization improved model performance by reducing the RMSE from 0.188203 to 0.115199. The best hyperparameter combination obtained was C = 100, epsilon = 0.01, and gamma = 0.1, achieving a cross-validation RMSE of 0.038695. The findings demonstrate that SVR with an RBF kernel optimized through Grid Search provides better predictive performance for natural rubber prices and has the potential to support decision-making processes for farmers, industry stakeholders, and policymakers.

Implementation of Gamified Mathematics Quizzes Using the Research and Development Method for Student Learning Evaluation

2026-08-15T17:48:48+07:00

Formal education is often supported by non-formal educational institutions, such as tutoring centers, to strengthen students’ academic understanding. However, conventional and monotonous paper-based evaluation processes are often perceived as stressful and burdensome by elementary school students, potentially reducing their interest in learning abstract basic mathematics concepts. This study aims to implement gamification concepts in an interactive mathematics quiz application to transform the evaluation process into a more engaging and enjoyable experience while increasing student engagement. The development method employed in this study is Research and Development (R&D) using the sequential Waterfall model. The application was developed using Flutter for the frontend and Firebase Cloud Firestore for the backend, incorporating key game elements such as Points and Leaderboards. Technical testing was conducted using the Black-Box Testing method, while product feasibility was assessed using a Likert-scale questionnaire. Improvements in learning interest and the effectiveness of learning outcomes were analyzed using the Normalized Gain (N-Gain Score) method based on pre-test and post-test results from 26 elementary school students. The feasibility evaluation conducted by experts and users yielded a score of 92.27%, which falls into the “Highly Feasible” category. Field testing results demonstrated an increase in the average score from 84.81 to 96.54, with an overall N-Gain index of 0.8256, categorized as “High” (Highly Effective). Therefore, the integration of gamification elements into a mobile application has been demonstrated to be valid and effective in improving students’ mathematics learning performance and engagement.

Internet of Things (IoT) and Telegram Based Palm Fruit Ripeness Detection System With K-Nearest Neighbor (K-NN) Algorithm

2026-08-17T23:29:43+07:00

Abstract

This research aims to overcome the difficulty in determining the level of maturity of oil palm fruit visually at PT Agro Muara Rupit by developing a more accurate fruit maturity detection system. This often results in suboptimal harvest results, there are some fruits that are harvested not always in a ripe state. As a result, yields are often not optimal, potentially causing losses for the company. To solve this problem, the research methods used include data collection through interviews, observations, and literature studies. The developed system is designed using a TCS3200 color sensor to detect fruit color, which is then processed by a NodeMCU microcontroller. RGB data obtained from the sensor will be calibrated and analyzed using the K- Nearest Neighbor algorithm to determine the level of ripeness based on the detected color sample data. The output of this system will be sent to the user via the Telegram application, the test results show that the developed system has an accuracy rate of 85% in detecting the maturity of oil palm fruit, with a value of K = 1. The application of Internet of Things (IoT) technology and the K-Nearest Neighbor algorithm is expected to increase efficiency and accuracy in detecting the maturity of oil palm fruit. Thus, this system will help optimize crop yields and reduce losses due to harvesting of immature fruit at PT Agro Muara Rupit.

Analisis Komparatif Decision Tree dan Random Forest untuk Klasifikasi Tingkat Obesitas

2026-08-17T23:16:06+07:00

Abstrak

Obesitas merupakan salah satu masalah kesehatan global yang terus mengalami peningkatan dan berkontribusi terhadap berbagai penyakit tidak menular, seperti diabetes melitus, hipertensi, dan penyakit kardiovaskular. Faktor gaya hidup, pola konsumsi makanan, serta aktivitas fisik memiliki peran penting dalam menentukan tingkat obesitas seseorang. Perkembangan teknologi machine learning memungkinkan pengembangan model prediksi yang dapat membantu mengidentifikasi tingkat obesitas berdasarkan karakteristik individu. Penelitian ini bertujuan untuk membandingkan kinerja algoritma Decision Tree dan Random Forest dalam klasifikasi tingkat obesitas berdasarkan faktor gaya hidup dan kondisi fisik. Dataset yang digunakan terdiri atas 2.111 data dengan 17 atribut yang mencakup informasi demografi, kebiasaan makan, aktivitas fisik, konsumsi air, penggunaan perangkat teknologi, serta riwayat keluarga. Tahapan penelitian meliputi prapemrosesan data, transformasi atribut kategorikal, pembagian data latih dan data uji, pelatihan model, serta evaluasi menggunakan metrik accuracy, precision, recall, dan F1-score. Hasil penelitian menunjukkan bahwa algoritma Random Forest memperoleh performa terbaik dengan nilai accuracy sebesar 95,27%, sedangkan Decision Tree memperoleh accuracy sebesar 91,73%. Selain itu, analisis feature importance menunjukkan bahwa atribut Weight, Height, Age, FCVC, dan Gender merupakan faktor yang memberikan kontribusi terbesar dalam proses klasifikasi tingkat obesitas. Hasil penelitian menunjukkan bahwa pendekatan ensemble learning pada Random Forest mampu meningkatkan kualitas klasifikasi dibandingkan model pohon keputusan tunggal.

Analysis of academic service information systems on the satisfaction of Sukarobot partners using the Servqual method

2026-08-13T00:17:42+07:00

The development of information technology has significantly impacted the improvement of service quality in the education sector, particularly in academic services. Sukarobot utilizes an academic service information system to support its service delivery to partners. However, several obstacles have been encountered in its implementation, such as occasional system errors, delayed service responses, and suboptimal service quality that meets user expectations. Therefore, this study aims to analyze the quality of the academic service information system and its impact on Sukarobot partner satisfaction using the SERVQUAL method.

This study employed a quantitative approach, with data collection techniques involving the distribution of questionnaires to Sukarobot partners. The SERVQUAL method measures service quality based on five dimensions: tangibles, reliability, responsiveness, assurance, and empathy. Data were analyzed using validity tests, reliability tests, Likert scale analysis, and gap analysis between user perceptions and expectations.

The results showed that all questionnaire items were valid, as their calculated r values were greater than the tabulated r values of 0.2133. Furthermore, the reliability test results showed that all variables had Cronbach's Alpha values above 0.60, thus declaring the research instrument reliable. Based on the SERVQUAL analysis, all dimensions had negative gap values, indicating that the service quality of Sukarobot's academic information system has not fully met user expectations. The assurance dimension had the smallest gap value of -0.34, while the reliability and responsiveness dimensions had the largest gap values of -0.40, making them a top priority for improvement

Design and Development of a Smart Flood Gate Prototype Based on Mamdani Fuzzy Logic and Multiple Sensors

2026-08-15T17:48:58+07:00

This study aims to design and implement a prototype of an adaptive automatic flood gate system based on Mamdani fuzzy logic using multi-sensor input, developed as part of a Capstone Design course. The system employs an ultrasonic sensor to measure water level, a rain sensor, and a temperature sensor as inputs, while a servo motor acts as the gate actuator. Mamdani fuzzy logic is applied to determine gate conditions classified into safe, alert, and danger levels. Experimental results show that the ultrasonic sensor achieves an average measurement error of ±1.2 cm compared to manual measurements. The system responds to water level changes with an average response time of 1.8 seconds. The flood gate operates correctly according to fuzzy rules with a 100% success rate across 15 test scenarios. Furthermore, the system demonstrates a decision accuracy of 93.3% in classifying flood conditions. The system can also be monitored in real time using the Blynk application.These results indicate that the proposed prototype performs effectively as a laboratory-scale flood gate control system and has potential for further development toward real-world implementation.

Sistem informasi tabungan dan manajemen kelompok kurban berbasis responsive web menggunakan metode rapid application development (RAD)

2026-08-10T23:30:55+07:00

The implementation of the qurban worship is often hampered by the readiness of community cash funds that must be paid in one lump sum close to the day of the event [1]. On the other hand, qurban managers at the mosque level generally still rely on manual recording based on physical books or spreadsheets, which has the potential to trigger recording errors (human error), data loss, and difficulties in mapping participant groupings (pooling) of 7 people per cow [1], [2]. This study aims to design and build a responsive web-based qurban savings information system that facilitates a periodic installment scheme and automates participant grouping [2], [3]. System development was carried out using the Rapid Application Development (RAD) method to accelerate the software engineering cycle through iteration and intensive user feedback [4], [5]. The test results show that the system built using responsive web design successfully makes it easier for the congregation to monitor savings balances and group status transparently through various devices, as well as assisting the committee in managing financial administration and group automation accurately [1], [6], [8].

Desain Model Jaringan IP Combat Management System pada Kapal Perang Republik Indonesia Menggunakan GNS3

2026-06-03T13:35:54+07:00

The modernization of the Primary Weapon Systems (Alutsista) on the Republic of Indonesia Warships (KRI) demands an integrated Combat Management System (CMS). The main problem faced is the shift of military communication systems toward Internet Protocol (IP)-based systems and Indonesia's dependence on foreign-made CMS technology. To overcome these problems, this research aims to design a new IP network architecture specifically tailored for KRI CMS operations and to test its functionality. The proposed solution is the design of a hybrid physical topology, combining the advantages of a star topology at the internal level of the Central Server Cabinet (CSC) and a partial mesh topology between cabinets. Additionally, data communication management is logically optimized using Virtual Local Area Networks (VLAN). The problem-solving method is conducted through modeling in the Graphical Network Simulator 3 (GNS3), followed by functional validation testing. The network simulation proves the seamlessness of comprehensive data exchange, indicated by successful inter-device connectivity across all test scenarios. The implementation of VLAN segmentation is proven successful in isolating data paths, ensuring that real-time tactical command flows can be distributed without being disrupted by high-capacity video loads. Overall, this network architecture design is proven to be robust and efficient in meeting operational standards that require redundancy, eliminating the risk of a single point of failure, and providing scalability for future upgrades. This architectural design is expected to serve as a solid blueprint for the domestic industry to support self-reliance in national defense technology.