Claim Missing Document
Check
Articles

Classification of Jombang Batik Motifs Using Ensemble Convolutional Neural Network Riza Satria Putra; Muhammad Muharrom Al Haromainy; Achmad Junaidi
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3204

Abstract

Batik, recognized by UNESCO as an Intangible Cultural Heritage, presents complex visual patterns that challenge automated classification systems. The intricate variations in texture, color, and geometry across motifs often lead to inconsistent performance in single Convolutional Neural Network (CNN) models, which struggle to generalize across subtle inter-class differences. To address these limitations, this study implements an Ensemble CNN framework to classify six Ploso Jombang batik motifs Garudan, Merak Kinasih Keyna Galeri, Ploso Bersemi, Jombang Berseri, Sulur Kangkung, and Burung Hong from a dataset of 2,134 images. The proposed approach integrates three pre-trained architectures EfficientNetB0, ResNet18, and VGG16 through a stacking ensemble strategy to leverage complementary feature extraction capabilities. Experimental results demonstrate that EfficientNetB0 achieved the highest individual accuracy (94%), while VGG16 recorded the lowest (60%). When combined, the ensemble configurations EfficientNetB0 + VGG16 and EfficientNetB0 + ResNet18 achieved peak test accuracies of approximately 96.88% on 321 test samples, reflecting a 2.88% improvement over the best single model. Confusion Matrix analysis confirmed robust model stability, with 100% accuracy for motifs such as Ploso Bersemi and Sulur Kangkung. These results validate that ensemble learning effectively mitigates overfitting and enhances generalization by aggregating diverse visual representations. The proposed model thus provides a reliable computational framework for automated batik classification and digital cultural preservation, supporting Indonesia’s efforts to document, catalog, and sustain its traditional heritage through artificial intelligence–driven methods.
Performance Evaluation of YOLOv5su and SVM With HOG Features for Student Attendance Face Recognition Achmad Rozy Priambodo; Achmad Junaidi; Muhammad Muharrom Al Haromainy
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3215

Abstract

The rapid evolution of Artificial Intelligence (AI) and Computer Vision has revolutionized conventional attendance systems by introducing automated and intelligent alternatives. Traditional approaches such as manual entry and fingerprint-based systems are often inefficient, error-prone, and unsuitable for large-scale student management. This study evaluates a hybrid face recognition framework that combines You Only Look Once version 5 su, Histogram of Oriented Gradients (HOG), and Support Vector Machine (SVM) to automate student attendance. The YOLOv5su algorithm performs fast and lightweight face detection, while HOG extracts gradient-based facial descriptors classified by SVM. Experiments were conducted using a facial image dataset consisting of 500 original images from 10 classes (50 images per class), which were augmented to 3,500 images with variations in pose, expression, and illumination. The proposed YOLOv5sU–HOG–SVM model achieved 97.1% detection accuracy and 97% recognition accuracy, with mean precision, recall, and F1-score values of 0.98, outperforming conventional CNN-based hybrid models in both accuracy and computational efficiency. These results demonstrate that the combination of YOLOv5su, HOG, and SVM provides a novel balance between detection speed and recognition robustness, making it suitable for real-time academic attendance management. Future work should integrate transformer-based facial feature extraction to further enhance robustness under extreme conditions and larger-scale datasets.
Mobile Legends Match Outcome Prediction Based on Players Statistics Using CatBoost and XGBoost Ciptaagung Firjat Ardine; Eka Prakarsa Mandyartha; Achmad Junaidi
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3259

Abstract

Mobile Legends: Bang Bang (MLBB) is a mobile-based Multiplayer Online Battle Arena (MOBA) game with a vast global community and professional ecosystem. Despite the extensive use of machine learning in desktop-based MOBAs such as Dota 2 and League of Legends, predictive modeling for MLBB remains underexplored. This study addresses this research gap by developing and comparing two advanced gradient boosting algorithms CatBoost and XGBoost for predicting match outcomes based on individual player statistics. The dataset, collected through web scraping from the official MPL Malaysia Season 14 website, comprises 1,430 player-level records representing professional-level competitive matches. Both models were trained and evaluated using 5-Fold Cross Validation to ensure stability and robustness. The results indicate that CatBoost achieved the highest predictive accuracy, with an average of 96.15%, outperforming XGBoost, which attained 94.75%. However, XGBoost exhibited exceptional computational efficiency, completing the prediction process 99.62% faster 0.76 seconds compared to CatBoost’s 3 minutes and 21 seconds. These findings highlight the trade-off between accuracy and processing speed in esports predictive modeling. The study demonstrates the potential of gradient boosting approaches for MLBB-specific analytics, providing a novel contribution to the limited body of research on mobile esports prediction. Accordingly, CatBoost is more suitable for analytical or strategic contexts where precision is essential, while XGBoost is better aligned with real-time predictive systems that demand rapid computation and scalability.
Autoimmune Skin Disease Image Classification using EfficientViT-M1 with AdamW Optimizer Hafiyan Fazagi Adnanto; Anggraini Puspita Sari; Achmad Junaidi
bit-Tech Vol. 8 No. 2 (2025): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i2.3300

Abstract

Diagnosing autoimmune skin diseases is a clinical challenge because several conditions share overlapping visual characteristics. This study evaluates the EfficientViT-M1 model trained with the AdamW optimizer to classify images from five autoimmune skin disease categories. The dataset contains 3,336 images before augmentation and is divided into 60 percent training, 20 percent validation, and 20 percent testing to ensure stable evaluation and reduce overfitting. The model is trained for 50 epochs with a learning rate of 0.0001, and experiments using batch sizes of 64, 128, and 256 are conducted to analyze the impact of data processing on performance. Performance is measured using accuracy, precision, recall, and F1-score derived from confusion matrix results. The best performance appears at a batch size of 64, achieving 89.25 percent accuracy along with balanced precision, recall, and F1-score. These results show that EfficientViT-M1 can extract relevant lesion features while maintaining computational efficiency. A notable challenge emerges when distinguishing visually similar disease classes, particularly Psoriasis and Lichen, which often share comparable textures and color patterns that contribute to misclassification. This highlights the influence of dataset imbalance and visual overlap on prediction outcomes. The approach offers potential value for clinical practice, especially in underserved areas where automated decision support can help early screening when specialist access is limited. The model demonstrates encouraging potential as a resource-efficient tool for dermatological assessment. Future improvements may include increasing dataset diversity, incorporating clinical metadata, and exploring alternative optimization strategies to enhance diagnostic reliability.
Klasifikasi Penyakit Mata Menggunakan ResNet-50 Berdasarkan Citra Fundus Kurniawan, Muh. Irsyad Dwi; Sari, Anggraini Puspita; Junaidi, Achmad
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3306

Abstract

Visual impairment from diabetic retinopathy, glaucoma, and cataracts remains a critical global health issue, emphasizing the need for early and accurate diagnosis to prevent permanent vision loss. This research presents an automated detection system utilizing ResNet-50, a deep learning architecture, to classify fundus images into multiple retinal disease categories. Unlike conventional convolutional neural networks used in prior studies, this approach leverages ResNet-50's residual learning mechanism to better identify complex retinal patterns. The study employed 4,184 fundus photographs from Kaggle, divided into four classes: cataract, diabetic retinopathy, glaucoma, and normal. Images were preprocessed through resizing to 224×224 pixels, normalized with ImageNet parameters, and augmented using random rotation and flipping techniques to enhance model generalization. Dataset splitting followed stratified sampling with an 80-20 train-test ratio, maintaining balanced class representation. Model training spanned 20 epochs using the Adam optimizer across three learning rates: 0.1, 0.01, and 0.001. The 0.001 learning rate produced optimal results with 90.35% accuracy, 90.28% precision, 90.18% recall, and 90.21% F1-score. The confusion matrix indicated strong performance in detecting diabetic retinopathy (219 correct predictions) and normal cases (189 correct predictions), though minor misclassifications occurred between glaucoma and normal categories. These findings validate ResNet-50's residual architecture as an effective tool for extracting discriminative retinal features, offering a computationally efficient solution for automated eye disease screening. Future work should incorporate explainability methods like Grad-CAM to enhance clinical interpretability and build trust among healthcare professionals in AI-assisted diagnostic systems.
Optimizing Gaussian Mixture Model Using Principal Component Analysis for Welfare Clustering Wahyu Gunawan, Rafif Ilafi; Al Haromainy, Muhammad Muharrom; Junaidi, Achmad
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3310

Abstract

Welfare inequality among regions remains a fundamental challenge in achieving balanced development across East Java Province. The complexity of social, economic, and development indicators often obscures the true patterns of regional welfare. To address this issue, this study proposes a more efficient analytical approach by integrating Principal Component Analysis (PCA) and the Gaussian Mixture Model (GMM) to cluster regions based on welfare levels. The dataset, obtained from the Central Bureau of Statistics (BPS) of East Java for the 2010–2024 period, includes diverse social and economic indicators. PCA was employed to reduce dimensionality and eliminate variable correlations, preserving the essential information within the data. The resulting principal components were then analyzed using GMM to uncover welfare clustering patterns. Based on the evaluation using the Bayesian Information Criterion (BIC) and silhouette score, the optimal configuration was achieved with two clusters, a tolerance of 1e-2, a maximum iteration of 200, and a silhouette score of 0.3403. The first cluster represented regions with higher welfare conditions, while the second indicated relatively lower welfare. These findings demonstrate that the PCA–GMM integration not only improves clustering accuracy but also enhances interpretability of welfare distribution across regions. Future studies may combine PCA with non-linear dimensionality reduction techniques such as Uniform Manifold Approximation and Projection (UMAP) to preserve local structures within complex datasets. Such integration is expected to reveal subtler and more dynamic welfare patterns, offering deeper insights into regional development disparities.
Development of Blockchain-Based Escrow System with IPFS Protocol for Secure Digital Transactions Sitompul, Pelean Alexander Jonas; Wahanani, Henni Endah; Junaidi, Achmad
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3337

Abstract

Digital transactions are essential to modern economic activities, yet challenges related to trust, transparency, and security persist. This research develops a blockchain-based escrow system integrated with the InterPlanetary File System (IPFS) to address these issues through a decentralized, tamper-resistant architecture. The primary aim is to create an escrow platform that minimizes human intervention while ensuring data integrity, thereby overcoming limitations found in traditional escrow mechanisms that rely heavily on legality and banking institutions. This study demonstrates the feasibility of blockchain technology enhancement to existing escrow models, especially for traders conducting high-value digital transactions. The system enables secure interactions between buyers, sellers, and viewers through a decentralized application (dApp) that assigns user roles and executes transaction logic. Funds are securely locked within the smart contract, while digital assets are stored in IPFS. In cases of dispute, the viewer can cancel the transaction, triggering an automated refund to the buyer and deletion of associated asset data to maintain fairness and security. Smart contract development and testing are carried out using the Hardhat framework before deployment to networks such as the Ethereum-based Sepolia Testnet. The results show that the proposed system reduces transaction risks, increases user trust, and enhances transparency throughout the digital transaction process. This research introduces a practical framework for decentralized escrow systems and provides valuable insights for industries seeking secure, blockchain-driven transaction solutions. The system developed in this study serves as a reference model for integrating traditional transaction with blockchain technology, encouraging broader adoption and future exploration of decentralized systems.
Comparative Analysis of LSTM and GRU Algorithms for Inflation Rate Forecasting Ardiyansyah, Moh. Angga; Al Haromainy, Muhammad Muharrom; Junaidi, Achmad
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3370

Abstract

Inflation is a critical economic indicator that directly affects price stability, purchasing power, and the formulation of fiscal and monetary policies. In East Java, inflation has demonstrated considerable year-to-year volatility, creating significant challenges for policymakers in maintaining regional economic stability. This situation highlights the need for forecasting models that are both accurate and capable of adapting to complex economic data patterns. This study presents a comparative analysis of two deep learning algorithms Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) for forecasting year-on-year (YoY) inflation in East Java using data from January 2005 to December 2024. The dataset was processed using Min–Max normalization and a 12-month sliding window to capture long-term dependencies and seasonal variations. Model performance was evaluated using RMSE, MAE, and MAPE. The findings demonstrate that no single model performs best across all metrics. The LSTM4 model with a [128,128] architecture achieved the lowest MAE and MAPE values, indicating superior average predictive accuracy and stronger capability in learning complex long-term inflation patterns. In contrast, the GRU1 [64,64] model produced the lowest RMSE and the shortest training time, highlighting its efficiency in minimizing extreme prediction errors and reducing computational cost. These results offer valuable insights for policymakers in East Java: LSTM is more suitable for applications requiring high prediction accuracy, whereas GRU is preferable for real-time or resource-efficient forecasting systems, especially in fast-changing economic environments.
Aplikasi OMR untuk Pemeriksaan Lembar Jawaban menggunakan DexiNed Prastyo, Kus Dwi; Junaidi, Achmad; Aditiawan, Firza Prima
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3425

Abstract

Digital image processing is a field of computer science that focuses on analyzing and interpreting digital images to extract meaningful information. One of its applications is Optical Mark Recognition (OMR), a technology used to detect marks on documents. OMR is commonly utilized for evaluating answer sheets. However, conventional OMR systems typically rely on specialized scanners that are expensive and lack flexibility. Although Computer-Based Testing (CBT) offers the convenience of automated scoring, its implementation heavily depends on the availability of technological infrastructure such as computers, internet connectivity, and a stable power supply. This study develops a real-time Optical Mark Recognition (OMR) application capable of performing answer sheet assessment directly on the client side. The system utilizes the DexiNed method for edge detection of the answer areas. The application is web-based and built using JavaScript and OpenCV.js to process images directly from the user's device camera. Testing was carried out under various scenarios, including different lighting intensities, scanner positions, pencil types, and shading quality. The results show that the application can detect marked answers with an accuracy up to 100%, although some limitations were observed under certain technical conditions. Weaknesses were found in low lighting conditions using a 5 watt lamp at a distance of 3 meters, light reflections, and the camera angle was not aligned with the answer sheet. Overall, the application provides an efficient and flexible alternative for answer sheet assessment without requiring dedicated scanning devices, making it suitable for educational institutions with limited infrastructure.
Design of Thesis Topic Recommendation System Using TF-IDF and Cosine Similarity Arrisalah, Muhammad Baihaqi; Haromainy, Muhammad Muharrom Al; Junaidi, Achmad
bit-Tech Vol. 8 No. 3 (2026): bit-Tech
Publisher : Komunitas Dosen Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32877/bt.v8i3.3579

Abstract

Selecting a thesis topic is a critical stage in a student’s academic journey and frequently poses substantial cognitive and procedural challenges. This study reports the design and implementation of the Computer Science Thesis Recommendation System (SRSIK Hub), a web-based decision-support platform aimed at improving the efficiency and accuracy of thesis topic selection. The primary novelty of this research lies in the systematic integration of Term Frequency–Inverse Document Frequency (TF-IDF) and Cosine Similarity within a large-scale academic corpus to model fine-grained semantic relevance between student interests and prior thesis documents, enabling more precise and transparent recommendations than conventional keyword-based searches. The system adopts a content-based filtering approach and processes approximately 4,000 thesis records collected from multiple university repositories. Textual data are preprocessed and transformed using TF-IDF vectorization, while Cosine Similarity is employed to rank candidate topics according to relevance. System effectiveness was evaluated using the WebUse Framework involving 75 student respondents. The evaluation yielded an overall score of 4.44 out of 5, indicating high usability, strong information quality, and reliable system functionality. This performance score demonstrates that the proposed recommendation model is not only technically sound but also practically applicable in real academic settings, where it can significantly reduce topic selection time and uncertainty for students. The results confirm that SRSIK Hub effectively supports students in identifying research topics aligned with their academic interests and competencies. Beyond local deployment, the system is transferable to other institutions for scalable thesis recommendation support.
Co-Authors Achmad Rozy Priambodo Afifudin, Muhammad Agung Mustika Rizki, Agung Mustika Akbar, Refansya Rachmad Akmal, Mohammad Faizal Al Fathoni, Hanif Andreas Nugroho Sihananto Andreas Nugroho Sihananto Anggraini Puspita Sari Anggraini Puspita Sari Anggraini Puspita Sari Ar Romandhon, Mitzaqon Gholizhan Ardiyansyah, Moh. Angga Arif Saifudin, Muhamad Ariq Musyaffah Ghufron, Althaf Arrisalah, Muhammad Baihaqi Bachtiar Riza Pratama Basuki Rahmat Basuki Rahmat Masdi Siduppa beni tiyas kristanti Ciptaagung Firjat Ardine Dafauzan Bilal Syaifulloh Darmawan, Marcellinus Aditya Vitro Diyasa, I Gede Susrama Mas Dunuroi Assuryani Dwi Arman Prasetya Efendi, Ridwan Eka Prakarsa Mandyartha Erik evranata Pardede Erik Iman Heri Ujianto Eva Yulia Puspaningrum Fatullah, Ryan Reynickha Fauzan Novriandy, Muhammad Fetty Tri Anggraeny Firza Prima Aditiawan Galan Ahmad Defanka Hafiyan Fazagi Adnanto Henni Endah Wahanani Henni Endah Wahanani I Gede Susrama Mas Diyasa Isworo, Muhamad Raihan Ramadhani Izzatul Fithriyah Kartini Kartini kristanti, beni tiyas Kurniawan, Muh. Irsyad Dwi Lesmana, Benedictus Rafael Mandyartha, Eka Prakarsa Maulana, Hendra Mochammad Yoga Firnanda Mohammad Haydir Awaludin Waskito Muhammad Azka Zaki Muhammad Muharrom Al Haromainy Mustika Rizki, Agung Mutiq Anisa Tanjung Muttaqin, Faisal Nugroho Sihananto, Andreas Nurlaili, Afina Lina Oktaviana, Dinda Friska Paramitha, Clara Diva Permanasari, Wahyu Melinda Prastyo, Kus Dwi Pratama, Novandi Kevin Prinafsika PW, Benar Setya Rachmadhany Iman Rafie Ishaq Maulana Rahmanda Putri, Endin Ratantja Kusumajati, Fatwa Rayya Ruwa'im Nafie Ridwan Efendi Riza Satria Putra Rizki, Agung Mustika Royan Fajar Sultoni Sajiwo, Achmad Fauzihan Bagus Salsabila, Belia Putri Sari, Allan Ruhui Fatmah Sebrina, Aida Fitriya Shahab, Muhammad Syaugi Sitompul, Pelean Alexander Jonas Syahbagus Radithya Haryo Santoso Thalita Syahlani Putri Tinambunan, Fernanda Vierino, Farrel Tiuraka Wahyu Gunawan, Rafif Ilafi Wardah Gracillaria Suharyono, Farra William Lijaya Therry, Renaldy Zaim, Mohammad Syarifuz