Jurnal PROCESSOR

Analisis Kinerja Algoritma K-Nearest Neighbor Dan Random Forest Untuk Deteksi Serangan Pada Jaringan Perangkat IoT

Muhammad Ilham Mansis — 2025-10-30

Deteksi serangan pada jaringan perangkat Internet of Things (IoT) menjadi tantangan penting dalam menjaga keamanan sistem yang semakin kompleks dan rentan terhadap ancaman siber. Sebagai upaya dalam mengatasi permasalahan tersebut, penelitian ini bertujuan untuk mengevaluasi kinerja algoritma K-Nearest Neighbor (KNN) dan Random Forest dalam mendeteksi berbagai jenis serangan pada jaringan perangkat IoT. Dataset yang digunakan adalah Aposemat IoT-23, yang berisi 1.446.599 entri data lalu lintas jaringan dari berbagai jenis serangan seperti Benign, DDoS, Attack, dan lainnya. Tahapan metode meliputi data preprocessing, data cleaning, label encoding, setelah itu dilakukan pelatihan model dan evaluasi menggunakan metrik accuracy, precision, recall, f1-score, ROC-AUC, serta validasi silang 5-Fold Cross-Validation. Hasil penelitian menunjukkan bahwa algoritma Random Forest memiliki kinerja lebih baik dibandingkan KNN, dengan F1-Macro Score sebesar 0,9396, ROC-AUC 0,9955, serta accuracy sebesar 92,20%. Sementara itu, KNN mencatatkan F1-Macro Score sebesar 0,9256, ROC-AUC 0,9867, dan accuracy sebesar 92,51%. Random Forest juga menunjukkan performa yang lebih stabil pada semua lipatan validasi silang. Berdasarkan temuan ini, Random Forest dinilai lebih efektif dalam mendeteksi serangan pada jaringan IoT.

Interactive Client-Side Network Simulation with Dynamic Bandwidth Management and Real-Time 3D Visualization

Lukman Lukman — 2025-10-30

Effective network management in dynamic environments necessitates robust tools for real-time analysis and visualization. This paper presents a client-side network simulation designed to model and visualize complex network behaviors, including dynamic bandwidth allocation, traffic spillover across multiple Internet Service Providers (ISPs), and system-wide congestion throttling. Leveraging Three.js, the system procedurally generates a 25-node network topology and simulates fluctuating traffic demands using a stochastic model. A central router implements a hierarchical spillover logic for external traffic and a proportional throttling algorithm to cap total bandwidth, ensuring network stability under peak loads. Real-time data visualization is achieved through animated data packets, color-coded node states, and an overlaid 2D time-series graph. Validation confirms the system's functional validity, demonstrating its potential as an educational tool and a proof-of-concept for sophisticated digital twins for network operations.

Sistem Pendataan Siswa Berbasis AI untuk Analisis Karakter dan Monitoring Perkembangan di SD IT Ananda Empat Lawang

Muhammad Sulkhan Nurfatih — 2025-10-30

Primary education plays a strategic role in shaping students’ character and competencies from an early age. However, many schools face challenges in managing student development data, particularly in aspects related to character, which are often subjective and difficult to quantify systematically. This study aims to develop an artificial intelligence (AI)-based student data system integrated with natural language processing (NLP) to support teachers in analyzing student character and monitoring individual growth. The research employed a Research and Development (R&D) approach consisting of four stages: needs analysis, system design, web application implementation, and system evaluation. The system integrates NLP, TF-IDF, and BERT for feature extraction, character classification using decision tree and ensemble learning, and ChatGPT API to generate character summaries and learning recommendations based on teachers’ observational texts. The application was developed as a web-based platform under the domain aidata.itananda.sch.id and was tested on 48 students at SD IT Ananda Empat Lawang. Evaluation results showed that the system achieved 88% classification accuracy and provided data visualizations such as character distribution graphs and a confusion matrix. This system is expected to improve teachers’ efficiency in understanding student character and designing more adaptive and data-driven learning strategies. The study supports digital transformation in primary education through the integration of AI into school information systems.

Early Detection of Microsleep in Four-Wheeled Vehicle Drivers Through Eye Images Using the YOLOv8 Algorithm

Rhadis Steffani Saputri — 2025-10-30

Microsleep adalah episode tidur singkat dan tiba-tiba yang terjadi tanpa kesadaran, terutama saat mengemudi, dan diketahui sebagai salah satu penyebab utama kecelakaan lalu lintas. Penelitian ini bertujuan untuk mengembangkan sistem deteksi dini untuk microsleep pada pengemudi kendaraan roda empat menggunakan algoritma YOLOv8. Sistem ini beroperasi dengan mendeteksi kondisi mata secara real-time melalui kamera dan mengeluarkan peringatan baik yang dapat didengar maupun dilihat ketika mata tetap tertutup selama lebih dari 3 detik. Dataset yang digunakan adalah SleepyDetect dari Roboflow Universe, yang terdiri dari dua kelas yang relevan: Mata Terbuka dan Mata Tertutup. Model dibor menggunakan YOLOv8n. Hasil evaluasi menunjukkan kinerja yang sangat baik, dengan presisi 97,2%, recall 96,7%, F1-score 0,97, dan mAP@0.5 sebesar 97,9%. Sistem ini berhasil diimplementasikan dan diuji dalam berbagai kondisi pencahayaan, menunjukkan respons deteksi yang cepat dan akurat. Penelitian ini menunjukkan bahwa kombinasi visi komputer dan YOLOv8 dapat diterapkan secara efektif untuk meningkatkan keselamatan pengemudi melalui deteksi mikrotidur dini.

Penerapan Metode YOLOv5 untuk Klasifikasi dan Deteksi Objek Menggunakan Video Non-Real-Time

Yovi Apridiansyah — 2025-10-30

Deteksi objek merupakan salah satu penerapan utama dari teknologi computer vision dalam bidang kecerdasan buatan. Salah satu algoritma deteksi objek yang banyak digunakan karena efisiensinya adalah YOLOv5. Penelitian ini bertujuan untuk menerapkan metode YOLOv5 dalam mendeteksi dan mengklasifikasikan objek kendaraan dan manusia pada rekaman video non-real-time di kawasan Simpang Tugu Pena Kota Bengkulu. Dataset yang digunakan merupakan video hasil dokumentasi lapangan, yang kemudian dianalisis menggunakan model YOLOv5 dengan pelatihan berbasis transfer learning. Untuk menjaga identitas objek antar-frame, sistem dikombinasikan dengan metode Kalman Filter dan SORT sebagai pelacak objek. Hasil pengujian menunjukkan bahwa model yang dibangun mampu melakukan deteksi objek dengan baik pada kondisi visual yang bervariasi, serta mencapai nilai akurasi yang tinggi berdasarkan pengukuran menggunakan matriks konfusi. Penelitian ini menunjukkan bahwa penerapan YOLOv5 efektif digunakan dalam sistem dokumentasi visual berbasis AI di lingkungan ruang publik yang dinamis.

Penerapan Metode Profile Matching pada Penilaian Kinerja Dosen

Geraldo Revanska Effendy — 2025-10-30

The evaluation of lecturer performance at Adzkia University faces challenges in terms of inefficient data processing. This research aims to implement the Profile Matching method to optimize the lecturer performance assessment system, evaluate its effectiveness, and develop an application based on this method. The research was conducted using a quantitative method employing Profile Matching, which includes several stages: GAP calculation, GAP mapping, core factor and secondary factor analysis, total value calculation, and ranking determination. The evaluation was conducted on 38 lecturers considering five main criteria: Adzkian Values, Education, Research, Community Service, and Supporting Activities, which are detailed in 28 sub-criteria. The implementation of the Profile Matching method proved to produce objective assessments by placing Lecturer 31 as the lecturer with the highest score (4.251), followed by Lecturer 3 and Lecturer 30 (4.092). The developed web-based application successfully integrated this method and improved the efficiency of the assessment process. This study demonstrates the effectiveness of the Profile Matching method in evaluating lecturer performance with more objective results. The implemented system helps BPSDM conduct assessments more efficiently and generate more structured reports.

Analisis Sentimen Terhadap Tagar Kabur Aja Dulu Di Twitter Menggunakan Metode Lexicon-Based

Eko Arip Winanto — 2025-10-30

Tagar #KaburAjaDulu sempat menjadi perbincangan hangat di media sosial Twitter, mencerminkan respons masyarakat digital Indonesia terhadap dinamika sosial dan politik yang sedang berlangsung. Penelitian ini bertujuan untuk mengevaluasi sentimen publik terhadap tagar tersebut dengan menerapkan pendekatan lexicon-based menggunakan InSet (Indonesia Sentiment Lexicon).Data penelitian diperoleh melalui teknik scraping dengan pustaka Tweet Harvest, menghasilkan 581 tweet berbahasa Indonesia yang memuat tagar #KaburAjaDulu. Analisis dilakukan menggunakan Google Colaboratory dengan dukungan pustaka Python. Tahapan penelitian mencakup pra-pemrosesan teks (pembersihan data, tokenisasi, stopword removal, serta stemming/lematisasi), klasifikasi sentimen dengan metode lexicon-based, dan visualisasi hasil.Hasil analisis menunjukkan bahwa sentimen negatif mendominasi dengan persentase 41,72%, diikuti sentimen netral sebesar 33,73% dan sentimen positif sebesar 24,55%. Kata-kata dominan pada kategori negatif merepresentasikan kritik, keluhan, dan sindiran yang banyak disampaikan dalam gaya bahasa satir khas media sosial. Temuan ini mengindikasikan bahwa tagar #KaburAjaDulu lebih sering digunakan sebagai sarana ekspresi ketidakpuasan publik terhadap kondisi sosial-politik nasional.Secara keseluruhan, pendekatan lexicon-based terbukti efektif dalam memberikan gambaran umum mengenai kecenderungan opini publik tanpa memerlukan pelatihan model. Namun, metode ini memiliki keterbatasan dalam menangkap makna kontekstual dari bahasa informal maupun sarkastik. Oleh karena itu, penelitian ini dapat dijadikan pijakan awal bagi studi lanjutan yang mengintegrasikan pendekatan machine learning untuk meningkatkan akurasi analisis sentimen pada media sosial.

Analisis Kepuasan Pengguna Aplikasi DPS-Denpasar Prama Sewaka Menggunakan Metode End User Computing Satisfaction (EUCS)

I Wayan Angga Arditaloka — 2025-10-30

The rapid advancement of information technology has encouraged government institutions to adopt digital systems to enhance the effectiveness and efficiency of public services. This study examines user satisfaction with the DPS-Denpasar Prama Sewaka application, which serves as a platform for public complaints and information services in Denpasar City. A quantitative approach was employed using Partial Least Squares Structural Equation Modeling (PLS-SEM), based on the End User Computing Satisfaction (EUCS) model. Data were collected from 100 randomly selected respondents. Five key variables were analyzed: content, accuracy, format, ease of use, and timeliness. The results indicate that collectively, these variables explain 75% of user satisfaction. However, only three variables-content (27.5%), ease of use (41.2%), and timeliness (29.7%)-have a significant individual impact. Accuracy and format were found to have no significant effect. These findings highlight the importance of improving information quality, accessibility, and prompt service delivery to enhance user satisfaction. The study provides practical recommendations for developers and local governments to refine digital public service systems that are more responsive and efficient in meeting community needs.

Design of Wireless Network Authentication System and File Server using RADIUS and WPA2-Enterprise

Tristanto Ari Aji Tristanto Ari Aji — 2025-10-30

Network security and efficiency are important aspects in managing network infrastructure, especially in wireless networks that are prone to various forms of attacks. This research aims to design a more secure and efficient wireless network and file server authentication system using the WPA2-Enterprise protocol and a RADIUS server. In this design, the WPA2-Enterprise protocol is used to improve the security of wireless network access with a user identity-based authentication method, not just a generic password. To improve the efficiency of the authentication process, a RADIUS server is used. The RADIUS (Remote Authentication Dial-In User Service) server functions as a centralised authentication centre that verifies user credentials from a predefined database. The system in this research integrates file server service authentication and wireless network usage authentication. Test results show that this design successfully provides more secure authentication, prevents unauthorised access, and simplifies user and access rights management. The use of a RADIUS server and the WPA2-Enterprise protocol allows for flexibility and efficiency in implementing the system in a campus environment. The implementation of this system can be a practical solution in improving the security and efficiency of the wireless network authentication process and internal file server services.

Voice-Based Depression Pattern Recognition Using Mel-Frequency Cepstral Coefficients Feature Extraction

Wahju Tjahjo Saputro — 2025-10-30

The identification of depression patterns from human voices is important because depression can interfere with activities, reduce interest in learning, and hinder socialisation. Depression is a significant problem today because there has been a global increase in the number of people suffering from it. The factors contributing to depression are numerous and complex, and can affect all groups, from children to the elderly. The purpose of this study was to identify depression patterns based on voice feature extraction. The feature extraction method used is Mel-Frequency Cepstral Coefficients (MFCC). The MFCC method is capable of extracting features that closely resemble the human auditory system. The dataset used is the EATD-Corpus, which contains 162 recordings of students from Tongji University in China. The results of the study show that depression and healthy patterns can be distinguished using MFCC parameters, namely 25 measurements per frame, 10 frame intervals, an alpha value of 0.97 as the pre-emphasis coefficient, a maximum of 40 Mel filterbank coefficients, and 12 cepstral coefficients. Classification thresholds can be obtained for two classes: healthy with thresholds < 53.00 and depressed ≥ 53.00 using the Self-Rating Depression Scale.

Analisis Performa, Overload, dan Kerentanan pada Website Basarnas Bengkulu untuk Optimalisasi Kinerja

Julyane Kevin Cheka — 2025-10-30

The Bengkulu Basarnas website plays a crucial role in providing information related to search and rescue activities to the public. However, with the increasing number of visitors and the need for fast and accurate information, the site faces several challenges related to performance, overload, and vulnerabilities such as slow page load times, a website that tends to be unresponsive, and indications of defacement, which are certainly detrimental. Therefore, research was conducted with the aim of analyzing and providing recommendations for optimizing website performance. Implementing the Software Testing Life Cycle (STLC) method, a method used for software testing, and assisted by tools such as PageSpeed Insights, Apache Jmeter, and Acunetix, enabled the research to be conducted comprehensively and efficiently. The results obtained an average performance score of 81.87, which is included in the good range but still needs improvement. On overload, the research results were obtained, namely scenario 1 with 5 users (105 samples) there was 0.00% error, scenario 2 with 15 users (315 samples) there was 0.32% error, and scenario 3 with 45 users (945 samples) there was 1.90% which overall shows that website performance is still quite good at low loads but decreases at high loads. And on vulnerabilities, the research results were in the Low severity vulnerabilities category, which means there are minor vulnerabilities and are still relatively safe but still need to be followed up for prevention efforts. Based on these findings, it is recommended to pay attention to server infrastructure, optimize content or images, and improve and pay attention to system security in order to minimize vulnerabilities to prevent threats and attacks that can disrupt website performance.