Garuda - Garba Rujukan Digital

Perbandingan Efficientnet, Visual Geometry Group 16, dan Residual Network 50 Untuk Klasifikasi Kendaraan Bermotor Andrianto, Andrianto; Tahyudin, Imam; Karyono, Giat
Building of Informatics, Technology and Science (BITS) Vol 6 No 3 (2024): December 2024
Publisher : Forum Kerjasama Pendidikan Tinggi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47065/bits.v6i3.6450

This study compares the performance of three Convolutional Neural Network (CNN) models—EfficientNet, VGG16, and ResNet50—in motor vehicle classification tasks using the "Car vs Bike" dataset. Transfer learning was applied using pretrained weights from ImageNet. The results indicate that VGG16 achieved the best performance with 95% accuracy, precision of 0.95, recall of 0.96, and an F1-score of 0.95, demonstrating high balance in recognizing both classes. ResNet50 attained 87% accuracy on the test dataset with a precision of 0.89, recall of 0.84, and an F1-score of 0.87, offering a trade-off between accuracy and computational efficiency. Conversely, EfficientNet exhibited the lowest performance with 50% accuracy, failing to recognize the "Car" class effectively, as evidenced by precision and recall values of 0.00. Factors such as architectural complexity, dataset bias, and computational efficiency influenced these outcomes. This study reinforces previous findings on the strengths and weaknesses of CNN models in motor vehicle classification applications. Furthermore, it highlights the importance of balanced data management and model selection tailored to specific application requirements. However, the dataset's limitation of only two classes and reliance on transfer learning remain areas for future improvement. These findings provide valuable insights for developing intelligent transportation systems requiring high accuracy and efficiency.

Penerapan Model Ensemble Learning dengan Random Forest dan Multi-Layer Perceptron untuk Prediksi Gempa Turino, Turino; Saputro, Rujianto Eko; Karyono, Giat
Jurnal Pendidikan dan Teknologi Indonesia Vol 5 No 2 (2025): JPTI - Februari 2025
Publisher : CV Infinite Corporation

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jpti.667

Penelitian ini mengusulkan model hybrid yang menggabungkan metode Random Forest (RF) dan Multi-Layer Perceptron (MLPRegressor) untuk memprediksi magnitudo gempa bumi. Model ini bertujuan untuk meningkatkan akurasi prediksi dengan memanfaatkan kekuatan kedua algoritma tersebut, yang masing-masing memiliki keunggulan dalam menangani hubungan non-linier dan mengenali pola kompleks dalam data seismik. Evaluasi model menggunakan tiga metrik utama, yaitu Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), dan R-squared (R²). Hasil evaluasi menunjukkan bahwa model hibrida ini mampu memprediksi magnitudo gempa dengan akurasi yang cukup baik, dengan MAE sebesar 0,0738, RMSE 0,1078, dan R² 0,4204. Penerapan praktis dari model ini sangat relevan untuk sistem peringatan dini gempa bumi yang dapat membantu masyarakat untuk mengambil langkah-langkah pencegahan, seperti evakuasi dan penguatan infrastruktur di wilayah yang berisiko tinggi. Penelitian ini juga membuka peluang untuk mengembangkan model lebih lanjut dengan memperkenalkan data seismik real-time, algoritma pembelajaran mesin yang lebih canggih, dan penggunaan data geofisik serta pengamatan satelit untuk meningkatkan akurasi prediksi gempa bumi di masa depan. Dengan terus melakukan inovasi, ada potensi untuk mengembangkan sistem prediksi gempa bumi yang lebih akurat dan dapat diandalkan, yang pada akhirnya dapat meningkatkan kesiapsiagaan dan ketahanan terhadap bencana alam.

Optimization of Recommender Systems for Image-Based Website Themes Using Transfer Learning Wahid, Arif Mu'amar; Hariguna, Taqwa; Karyono, Giat
Journal of Applied Data Sciences Vol 6, No 2: MAY 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i2.671

Recommender systems play a crucial role in personalizing user experiences in e-commerce, digital media, and web design. However, traditional methods such as Collaborative Filtering and Content-Based Filtering struggle to account for visual preferences, limiting their effectiveness in domains were aesthetics influence decision-making, such as website theme recommendations. These systems face challenges such as data sparsity, cold-start problems, and an inability to capture intricate visual features. To address these limitations, this study integrates Convolutional Neural Networks (CNNs) with advanced recommendation models, including Inception V3, DeepStyle, and Visual Neural Personalized Ranking (VNPR), to enhance the accuracy and personalization of visually-aware recommender systems. A quantitative research approach was employed, using controlled experiments to evaluate different combinations of feature extractors and recommendation models. Data was sourced from ThemeForest, a widely used platform for website themes, and underwent preprocessing to ensure consistency. The models were evaluated using precision, recall, F1 score, Mean Average Precision (MAP), and Normalized Discounted Cumulative Gain (NDCG) to measure recommendation quality. The results indicate that Inception V3 + VNPR outperforms other model combinations, achieving the highest accuracy in personalized theme recommendations. The integration of transfer learning further improved feature extraction and performance, even with limited training data. These findings underscore the importance of combining deep learning-based feature extraction with recommendation models to improve visually-driven recommendations. This study provides a comparative analysis of CNN-based recommender systems and contributes insights for optimizing recommendations in visually complex domains. Despite improvements, challenges such as dataset diversity remain a limitation, affecting generalizability. Future research could explore alternative CNN architectures, such as ResNet and DenseNet, and incorporate user feedback mechanisms to further enhance recommendation accuracy and adaptability.

Penerapan CNN dan RNN untuk Pembuatan Deskripsi Konten Visual Menggunakan Deep Learning Hermanto, Aldy Agil; Karyono, Giat; Tahyudin, Imam
Building of Informatics, Technology and Science (BITS) Vol 6 No 4 (2025): March 2025
Publisher : Forum Kerjasama Pendidikan Tinggi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47065/bits.v6i4.6958

The development of technology in the field of image and sound processing has had a significant impact on increasing the accessibility of information for various groups, especially for individuals with visual impairments. One of the innovations that emerged was the image to speech system, which allows the conversion of images into sounds that can be understood by its users. The main problem lies in the low accuracy of object recognition in images with high variability, such as poor lighting or complex backgrounds, as well as the challenge of producing suitable text descriptions to be converted into audio. The method used involves extracting image features using InceptionV3-based CNN and forming a sequence of descriptive texts through RNN with an attention mechanism. The dataset consists of 40,455 captions and 8,091 images, processed using text and image pre-processing techniques before being trained using the teacher forcing technique. The evaluation results show a very low BLEU score (5.154827976372712e-153), indicating the model's inability to replicate the original caption well. However, the audio from the text-to-speech conversion using Google Text-to-Speech is quite clear. Future solutions include increasing the dataset, applying regularization, and adjusting the model architecture to improve the accuracy of caption prediction and audio relevance to the image. With these improvements, it is hoped that the system can provide more inclusive visual information accessibility for individuals with visual impairments.

Pengembangan Aplikasi Informasi Kos Mahasiswa dengan Metode RAD di Universitas Amikom Purwokerto Ibrahim, Farrel; Karyono, Giat; Purwadi; Nurfaizal, Yusmedi
Jurnal Pendidikan dan Teknologi Indonesia Vol 5 No 9 (2025): JPTI - September 2025
Publisher : CV Infinite Corporation

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jpti.979

Keterbatasan informasi mengenai tempat kos di sekitar Universitas Amikom Purwokerto menjadi tantangan bagi mahasiswa dalam menemukan hunian yang sesuai. Permasalahan ini mendorong perlunya solusi teknologi yang efisien, akurat, dan mudah diakses. Penelitian ini bertujuan untuk merancang dan mengembangkan aplikasi mobile berbasis Android yang menyajikan informasi kos secara lengkap dan terintegrasi. Metode pengembangan yang digunakan adalah Rapid Application Development (RAD) dengan fitur utama berupa pencarian kos, detail informasi, integrasi Google Maps, serta komunikasi langsung melalui WhatsApp dengan pemilik kos. Uji coba aplikasi dilakukan terhadap 47 mahasiswa pengguna melalui User Acceptance Testing. Hasil evaluasi menunjukkan bahwa 95,7% responden menyatakan aplikasi mudah digunakan, 91,4% menyatakan fitur sesuai dengan kebutuhan, dan 93,6% menyatakan informasi dalam aplikasi jelas dan mudah dipahami. Instrumen survei juga telah diuji validitas dan reliabilitasnya, dengan nilai Alpha Cronbach sebesar 0,822 yang menunjukkan bahwa instrumen tergolong reliabel. Penelitian ini memberikan kontribusi dalam pengembangan sistem informasi berbasis mobile yang mendukung kebutuhan mahasiswa dalam pencarian tempat tinggal. Aplikasi ini juga berpotensi diadaptasi secara lebih luas untuk meningkatkan efisiensi pencarian informasi hunian di lingkungan pendidikan tinggi di Indonesia.

Enhancing Customer Purchase Behavior Prediction Using PSO-Tuned Ensemble Machine Learning Models Kafilla, Princess Iqlima; Utomo, Fandy Setyo; Karyono, Giat
Jurnal Teknik Informatika (Jutif) Vol. 6 No. 5 (2025): JUTIF Volume 6, Number 5, Oktober 2025
Publisher : Informatika, Universitas Jenderal Soedirman

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jutif.2025.6.5.4952

Predicting customer purchase behavior remains a significant challenge in e-commerce and marketing analytics due to its complex and nonlinear patterns. This study introduces a machine learning framework that integrates ensemble learning models with Particle Swarm Optimization (PSO) for hyperparameter tuning to improve classification accuracy and class discrimination. Several ensemble algorithms, including CatBoost, XGBoost, LightGBM, AdaBoost, and Gradient Boosting, were compared against a baseline Logistic Regression model, both with default and PSO-optimized configurations. Experiments on a real-world e-commerce dataset containing behavioral and demographic variables showed that ensemble methods substantially outperformed traditional models across accuracy, F1-score, and ROC AUC metrics. Notably, the PSO-tuned Gradient Boosting model achieved the highest ROC AUC of 0.9547, improving the AUC by approximately 0.0076 compared to its default configuration, while CatBoost obtained the highest overall accuracy and F1-score. PSO optimization was especially effective in enhancing simpler models such as Logistic Regression but showed marginal gains and some convergence instability in more complex ensemble models. Feature importance analyses consistently identified variables such as time spent on the website, discounts availed, age, and income as key drivers of purchase intent. These findings demonstrate the benefit of combining ensemble learning with metaheuristic optimization, offering actionable insights for developing robust, data-driven marketing strategies.

Comparative Analysis of Decision Tree, Random Forest, Svm, and Neural Network Models for Predicting Earthquake Magnitude Turino, Turino; Saputro, Rujianto Eko; Karyono, Giat
Jurnal Teknik Informatika (Jutif) Vol. 6 No. 2 (2025): JUTIF Volume 6, Number 2, April 2025
Publisher : Informatika, Universitas Jenderal Soedirman

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jutif.2025.6.2.2378

This study conducts a comparative analysis of four machine learning algorithms—Decision Tree, Random Forest, Support Vector Machine (SVM), and Neural Network—to predict earthquake magnitudes using the United States Geological Survey (USGS) earthquake dataset. The analysis evaluates each model's performance based on key metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The Random Forest model demonstrated superior performance, achieving the lowest MAE (0.217051), lowest RMSE (0.322398), and highest R² (0.574261), indicating its robustness in capturing complex, non-linear relationships in seismic data. SVM also showed strong performance, with competitive accuracy and robustness. Decision Tree and Neural Network models, while useful, had comparatively higher error rates and lower R² values. The study highlights the potential of ensemble learning and kernel methods in enhancing earthquake magnitude prediction accuracy. Practical implications of the findings include the integration of these models into early warning systems, urban planning, and the insurance industry for better risk assessment and management. Despite the promising results, the study acknowledges limitations such as reliance on historical data and the computational intensity of certain models. Future research is suggested to explore additional data sources, advanced machine learning techniques, and more efficient algorithms to further improve predictive capabilities. By providing a comprehensive evaluation of these models, this research contributes valuable insights into the effectiveness of various machine learning techniques for earthquake prediction, guiding future efforts to develop more accurate and reliable predictive models.

Analisis Sentimen E-Commerce dengan Optimasi IndoBERT Menggunakan Bayesian Optuna Putri, Dewi Kartika; Karyono, Giat; Tahyudin, Imam
Jurnal Pendidikan dan Teknologi Indonesia Vol 5 No 12 (2025): JPTI - Desember 2025
Publisher : CV Infinite Corporation

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jpti.1208

Pertumbuhan e-commerce di Indonesia memicu peningkatan signifikan jumlah ulasan pengguna terhadap aplikasi Shopee di Google Play, sehingga analisis manual menjadi tidak lagi efisien. Penelitian ini bertujuan untuk meningkatkan performa analisis sentimen pada ulasan pengguna e-commerce berbahasa Indonesia dengan mengoptimalkan model IndoBERT menggunakan teknik Bayesian Optimization melalui framework Optuna. Latar belakang penelitian ini berfokus pada pentingnya pengolahan opini konsumen di platform Shopee, yang semakin tidak dapat ditangani secara manual karena volume data yang besar dan keberagaman gaya bahasa. Metode yang digunakan meliputi pengumpulan data melalui web scraping, praproses teks, pembagian data, pemodelan dengan IndoBERT, serta penalaan hiperparameter menggunakan Optuna. Hasil eksperimen menunjukkan bahwa optimasi Bayesian mampu meningkatkan akurasi klasifikasi dari 89,30% menjadi 96,10% dan macro-F1 dari 85,83% menjadi 94,82%. Selain itu, false-positive dan false-negative masing-masing turun sebesar 60% dan 67%, serta nilai ROC-AUC meningkat signifikan dari 0,9028 menjadi 0,9903. Temuan ini menegaskan efektivitas Optuna dalam meningkatkan performa dan efisiensi sistem klasifikasi sentimen berbasis IndoBERT, yang dapat diintegrasikan dalam pemantauan opini secara real-time di ranah e-commerce.

Pengaruh Prediksi Kredit bermasalah Terhadap Keputusan Bank Dengan Komparasi Metode ANN, CNN, dan Random Forest Nurmalitasari, Gupita; Eko Saputro, Rujianto; Karyono, Giat
Jurnal Pendidikan dan Teknologi Indonesia Vol 5 No 12 (2025): JPTI - Desember 2025
Publisher : CV Infinite Corporation

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52436/1.jpti.1276

Kredit bermasalah menjadi salah satu tantangan utama sektor perbankan karena dapat mengganggu stabilitas keuangan dan mempengaruhi keputusan strategis. Oleh karena itu, kemampuan memprediksi potensi kredit bermasalah secara akurat sangat penting dalam pengelolaan risiko. Penelitian ini membandingkan kinerja tiga algoritma kecerdasan buatan ANN (Artificial Neural Network) dengan arsitektur MLP (Multilayer Perceptron), CNN (Convolutional Neural Network), dan RF (Random Forest) untuk klasifikasi risiko kredit berdasarkan data nominatif kredit.Dataset mencakup fitur-fitur utama seperti plafon pinjaman, jangka waktu, penghasilan, baki debet, kolektibilitas dan total tunggakan. Data diklasifikasikan dalam lima kategori risiko berdasarkan status pembayaran antara lain L (Lancar), DP (Dalam Perhatian), KL (Kurang Lancar), D (Diragukan), dan M (Macet). Setiap model dievaluasi berdasarkan akurasi keseluruhan dan kemampuannya mengidentifikasi kelas minoritas dan mayoritas.Hasil menunjukkan bahwa Random Forest unggul dengan akurasi 91,42%, efektif dalam mengklasifikasikan kategori “Lancar” dan “Macet”. Model ANN MLP mencapai akurasi 89,90%, namun kurang optimal untuk kelas minoritas. Sementara CNN hanya mencatat akurasi 58,58% dan mengalami overfitting terhadap kelas mayoritas.Studi ini menyimpulkan bahwa Random Forest adalah metode paling tepat untuk data tabular yang seimbang, sementara CNN memiliki potensi dalam konteks data spasial atau sekuensial. Dan hasil yang akurat dari prediksi Random forest diharapkan dapat menjadi pertimbangan bagi lembaga keuangan dalam pengambilan keputusan terkait kebijakan di bank Pemerintah di Purbalingga.

Pemodelan Literasi TIK dan Kualitas Sistem untuk Memprediksi Kepuasan Pengguna: Peran Mediasi Kegunaan dan Moderasi Kesiapan Digital Ibrahim, Farrel; Karyono, Giat; Karyono, Purwadi; Nurfaizal, Yusmedi
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 13 No 1: Februari 2026
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2026131

Di era transformasi digital yang berlangsung cepat, kepuasan pengguna terhadap sistem teknologi tidak lagi semata-mata ditentukan oleh kinerja teknis, tetapi juga oleh kesiapan kognitif dan kontekstual pengguna. Penelitian ini mengkaji hubungan antara literasi TIK, kualitas sistem, persepsi kegunaan, dan kepuasan pengguna, serta menelaah peran mediasi dari persepsi kegunaan dan moderasi dari kesiapan digital. Pendekatan kuantitatif digunakan melalui pemodelan persamaan struktural Partial Least Squares (PLS-SEM) berdasarkan data dari 100 pengguna platform layanan digital. Hasil menunjukkan bahwa literasi TIK dan kualitas sistem secara signifikan meningkatkan persepsi kegunaan dan langsung memengaruhi kepuasan pengguna. Persepsi kegunaan juga terbukti memediasi pengaruh literasi TIK dan kualitas sistem terhadap kepuasan. Selain itu, kesiapan digital memoderasi secara positif pengaruh persepsi kegunaan terhadap kepuasan, menunjukkan bahwa pengguna dengan kesiapan lebih tinggi memperoleh manfaat lebih besar dari sistem yang dianggap berguna. Temuan ini memperkuat pengembangan model TAM melalui integrasi dimensi kognitif, teknologi, dan kontekstual, serta memberikan implikasi praktis bagi desain sistem akademik digital yang responsif terhadap kesiapan pengguna. Abstract In the era of rapid digital transformation, user satisfaction with technological systems is no longer solely determined by technical performance, but also by users’ cognitive and contextual readiness. This study examines the relationships among ICT literacy, system quality, perceived usefulness, and user satisfaction, while exploring the mediating role of perceived usefulness and the moderating role of digital readiness. A quantitative approach was employed using Partial Least Squares Structural Equation Modeling (PLS-SEM) based on data from 100 users of digital service platforms. The results indicate that ICT literacy and system quality significantly enhance perceived usefulness and directly influence user satisfaction. Perceived usefulness was also found to mediate the effects of ICT literacy and system quality on satisfaction. Moreover, digital readiness positively moderates the influence of perceived usefulness on user satisfaction, suggesting that users with higher readiness levels gain greater benefits from systems they perceive as useful. These findings strengthen the development of the TAM framework through the integration of cognitive, technological, and contextual dimensions, while also offering practical implications for the design of academic digital systems that are responsive to users’ readiness levels.

Title

Found 10 Documents
Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title Search

Found 10 Documents Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title

Found 10 Documents
Search