Claim Missing Document
Check
Articles

Found 30 Documents
Search

IMPLEMENTATION OF WORD EMBEDDING IN DETECTING POLITICAL FAKE NEWS IN INDONESIA USING LONG SHORT-TERM MEMORY ALGORITHM Anggit Rianansyah; Ema Utami; Dhani Ariatmanto
JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika) Vol 10, No 4 (2025)
Publisher : STKIP PGRI Tulungagung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29100/jipi.v10i4.6680

Abstract

This research investigates the use of word embedding techniques to detect political fake news in Indonesia by utilizing the Long Short-Term Memory (LSTM) algorithm. The spread of fake news, particularly in the political realm, poses significant challenges to public trust and the integrity of information. To address these challenges, we employed a dataset of political news articles and applied word embedding to convert the text into a numerical format that represents the semantic relationships between words. The LSTM algorithm, known for its ability to process and learn from sequential data, was then used to identify patterns indicative of fake news. Our model demonstrated satisfactory accuracy, with the LSTM algorithm without word embedding achieving an average accuracy of 67%, while the application of word embedding (Word2Vec, Glove, and FastText) resulted in average accuracies of 84%, 81%, and 86%, respectively. These findings confirm that combining word embedding with LSTM is effective in detecting fake news. This research contributes to ongoing efforts to combat misinformation in Indonesia by providing a robust tool to enhance the reliability of news in the digital age. Further developments, such as the integration of additional linguistic features and the expansion of the dataset, are expected to improve the model’s performance and adaptability across various contexts.
Evaluasi Pengalaman Pengguna Aplikasi Presensi Digital SIHADIR Polnep Menggunakan User Experience Questionnaire Plus (UEQ+) Muhammad Ghozy Alkhairi; Dhani Ariatmanto
JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika) Vol 11, No 2 (2026)
Publisher : STKIP PGRI Tulungagung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29100/jipi.v%vi%i.10407

Abstract

Transformasi digital di perguruan tinggi menuntut sistem informasi yang fungsional sekaligus memberikan pengalaman pengguna (User Experience/UX) yang baik. SIHADIR Polnep adalah aplikasi presensi digital berbasis Android yang diterapkan di Politeknik Negeri Pontianak sejak Januari 2025. Namun, pasca-implementasi, ditemukan indikasi permasalahan pada antarmuka yang kurang intuitif dan kendala teknis yang mengganggu kenyamanan pengguna. Penelitian ini bertujuan menganalisis tingkat pengalaman pengguna aplikasi SIHADIR menggunakan User Experience Questionnaire Plus (UEQ+) dengan empat dimensi: Perspicuity, Efficiency, Dependability, dan Attractiveness. Kuesioner disebarkan kepada 82 pengguna aktif yang terdiri dari dosen dan tenaga kependidikan Polnep. Hasil menunjukkan Attractiveness (mean=1,45) dan Efficiency(mean=1,43) berkategori Positif, sedangkan Perspicuity (mean=0,698) dan Dependability (mean=0,256) berkategori Netral. Uji Mann-Whitney U mengungkap perbedaan persepsi signifikan antara kedua kelompok (p0,001), paling mencolok pada dimensi Dependability, yang dijelaskan melalui konsep konsekuensi penggunaan (stakes of use). Temuan ini menjadi acuan kuantitatif pengalaman sekaligus landasan rekomendasi perbaikan antarmuka berbasis data.
Comparative Analysis of Live Action Film Production Management Using Critical Path Method (CPM) Versus Conventional Production Processes Agung Nugroho; Mohammad Suyanto; Dhani Ariatmanto
Intechno Journal : Information Technology Journal Vol. 7 No. 1 (2025): July
Publisher : Universitas AMIKOM Yogyakarta

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24076/intechnojournal.2025v7i1.2019

Abstract

The production of the film “Kinah dan Redjo”, by Universitas Amikom and MSV Sinema, has been completed, prompting researchers to conduct an analysis and evaluation of the production management applied. The focus of this study is on time and cost, which are critical factors supporting film production. An extended production duration was identified as a challenge, as it reduces effectiveness and leads to cost overruns. Therefore, this study aims to compare project management strategies for successful planning and control, using both conventional methods and the Critical Path Method (CPM). This analysis is expected to yield faster project completion and establish efficient, productive standards for future productions. The conventional approach indicated a total production duration of 681 days, comprising 120 days for pre-production, 18 days for production, and 551 days for post-production. Upon analysis using the CPM method, the total duration was reduced to 459 days, including 113 days for pre-production, 152 days for production, and 191 days for post-production. The graphical comparison of methods shows significant cost fluctuations across each production phase with the conventional method, especially increased costs during production despite the shorter duration. Conversely, the CPM method demonstrates more controlled and measurable durations and costs. This study underscores the importance of cost optimization, standardization of the Work Breakdown Structure (WBS), and hybrid modeling to enhance efficiency in dynamic film projects. Furthermore, this analysis serves as a foundational reference for the architectural planning of future applications incorporating artificial intelligence (AI) integration. AI has the potential to accelerate scheduling, optimize resource allocation, and streamline cost management and production design, thereby improving overall project efficiency.
OVERFITTING MITIGATION AND TRAINING OPTIMIZATION STRATEGIES IN DEEP LEARNING-BASED IMAGE CLASSIFICATION: A SYSTEMATIC REVIEW Budiman Budiman; Arief Setyanto; Andi Sunyoto; Dhani Ariatmanto
JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer) Vol. 12 No. 1 (2026): JITK Issue August 2026
Publisher : LPPM Nusa Mandiri

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33480/jitk.v12i1.7948

Abstract

Overfitting remains a critical challenge in developing deep learning–based image classification models, particularly as modern architectures become increasingly complex and parameter-intensive. Although convolutional and transformer-based models have demonstrated strong predictive performance, their generalization  capability depends strongly on the availability of large, well-annotated, and diverse training datasets. This condition is often difficult to achieve in real-world domains such as medicine, agriculture, and industrial inspection. Previous survey studies have examined various techniques related to image classification, data augmentation, and optimization strategies; however, these studies typically analyse individual approaches in isolation. As a result, opportunities remain to further synthesise the relationships among the underlying causes of overfitting, mitigation strategies, and training parameter configurations within a unified analytical perspective. This study employed the PRISMA framework to conduct a Systematic Literature Review (SLR). A total of 174 primary studies published between 2020 and 2025 were traced to address the gap. The review identifies three major sources of overfitting in image classification tasks: limited labeled data, model architecture complexity, and data and label quality issues. Based on these findings, the study synthesises the corresponding mitigation strategies reported in the literature. The main contribution of this study is not the proposal of a new theoretical taxonomy, but the systematic organisation and synthesis of methodological evidence into an integrated analytical framework. The resulting analytical framework relates the underlying causes of overfitting, mitigation strategies, training optimisation mechanisms, and parameter configuration practices within a unified perspective, thereby facilitating a more integrated interpretation of how these complementary aspects contribute to model generalisation in deep learning–based image classification.
FORECASTING STOCK MARKET MODEL: A SYSTEMATIC LITERATURE REVIEW Elia Setiana; Kusrini Kusrini; Tonny Hidayat; Dhani Ariatmanto
JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer) Vol. 12 No. 1 (2026): JITK Issue August 2026
Publisher : LPPM Nusa Mandiri

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33480/jitk.v12i1.7991

Abstract

The increasing digitisation of stock markets and the growing diversity of financial data sources have intensified the need for accurate, robust, and risk-aware stock market forecasting. This systematic literature review synthesises recent evidence to examine the effectiveness of forecasting methods under different data and market conditions, the characteristics of commonly used benchmark datasets, the contribution of preprocessing strategies, and the evaluation and validation practices applied in stock market forecasting. Following the PRISMA framework, 71 peer-reviewed studies retrieved from the Scopus database were systematically screened, classified, and analysed. The evidence mapping shows that sequence-based deep learning models, including LSTM, GRU, and CNN–LSTM, represent the largest methodological group at approximately 41%, followed by transformer- and attention-based approaches at around 16%. Volatility-oriented econometric and classical statistical models account for approximately 18% and 14%, respectively, while probabilistic and quantile-based approaches remain limited. The findings indicate that forecasting performance is strongly context-dependent: classical models remain effective for relatively stationary univariate series, volatility-oriented models are particularly relevant when clustering and spillover effects are present, and deep learning and transformer-based approaches are more suitable for multivariate, nonlinear, and feature-rich settings. Overall, the review highlights the need for greater integration of uncertainty-aware evaluation, regime-sensitive validation, and risk-oriented forecasting frameworks.
Analysis and Accuracy Improvement of Deep Learning Models for Ai-Generated Image Detection Otniel Putra Zebua; Dhani Ariatmanto
Jurnal Teknologi Informatika dan Komputer Vol. 12 No. 2 (2026): Jurnal Teknologi Informatika dan Komputer
Publisher : Universitas Mohammad Husni Thamrin

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.37012/jtik.v12i2.4025

Abstract

The rapid development of generative artificial intelligence makes authentic photographs increasingly difficult to distinguish from AI-generated images. This study analyzes EfficientNet-B3 and improves the detector using a Convolutional Block Attention Module (CBAM). A balanced subset of 5,000 CIFAKE images was used, containing 2,500 REAL and 2,500 AI-generated images. After duplicate bytes were removed, 4,000 images were allocated to stratified five-fold cross-validation and 1,000 images were retained as a held-out test set. Images were resized to 300×300 pixels and processed using gray-world color normalization, contrast enhancement, and training augmentation. The baseline achieved 88.25±1.20% validation accuracy and an 88.28±1.14% F1-score, whereas the initial CBAM configuration achieved 87.52±0.90% and 87.49±0.94%. However, the five-checkpoint CBAM ensemble reached 89.00% test accuracy and an 88.98% F1-score, improvements of 0.80 and 0.85 percentage points over the baseline ensemble. Ablation showed that CBAM-last achieved an 88.32% validation F1-score. The findings indicate that appropriate attention placement and checkpoint ensembling can improve performance; external test sets are recommended to assess generalization. The rapid development of generative artificial intelligence makes authentic photographs increasingly difficult to distinguish from AI-generated images. This study analyzes EfficientNet-B3 and improves the detector using a Convolutional Block Attention Module (CBAM). A balanced subset of 5,000 CIFAKE images was used, containing 2,500 REAL and 2,500 AI-generated images. After duplicate bytes were removed, 4,000 images were allocated to stratified five-fold cross-validation and 1,000 images were retained as a held-out test set. Images were resized to 300×300 pixels and processed using gray-world color normalization, contrast enhancement, and training augmentation. The baseline achieved 88.25±1.20% validation accuracy and an 88.28±1.14% F1-score, whereas the initial CBAM configuration achieved 87.52±0.90% and 87.49±0.94%. However, the five-checkpoint CBAM ensemble reached 89.00% test accuracy and an 88.98% F1-score, improvements of 0.80 and 0.85 percentage points over the baseline ensemble. Ablation showed that CBAM-last achieved an 88.32% validation F1-score. The findings indicate that appropriate attention placement and checkpoint ensembling can improve performance; external test sets are recommended to assess generalization.
Perbandingan Metode Ekstraksi Fitur LBP, GLCM, dan Canny dalam Klasifikasi Penyakit Daun Padi dengan KNN Roy Jordy; Dhani Ariatmanto
Bangkit Indonesia Vol. 14 No. 2 (2025): (April-October) Bangkit Indonesia
Publisher : LPPM STT Indonesia Tanjung Pinang

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Accurate and timely identification of rice leaf diseases plays a crucial role in supporting early disease control efforts in agriculture. This study aims to compare the performance of three image feature extraction methods—Local Binary Pattern (LBP), Gray Level Co-occurrence Matrix (GLCM), and Canny Edge Detection—in classifying three types of rice leaf diseases: Bacterial leaf blight, Brown spot, and Leaf smut. Each method was evaluated based on its confusion matrix as well as key performance metrics, including accuracy, precision, recall, and F1-score. Experimental results show that LBP achieved the highest classification performance with an accuracy of 92.06%, followed by GLCM at 78.57% and Canny at 66.67%. In addition to accuracy, LBP also outperformed the other methods across all evaluation metrics. These findings indicate that the local texture features captured by LBP are more effective in distinguishing disease types compared to the global texture features from GLCM and edge-based features from Canny. Therefore, LBP is recommended as a superior feature extraction method for automated classification systems of rice leaf diseases based on digital imagery.
Analisis Perbandingan Kinerja Model YOLO11 dan YOLOv8 dalam Identifikasi Penyakit pada Daun Tomat Muhammad Arif Kholis Majid; Dhani Ariatmanto
Bangkit Indonesia Vol. 14 No. 2 (2025): (April-October) Bangkit Indonesia
Publisher : LPPM STT Indonesia Tanjung Pinang

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Diseases on tomato leaves can reduce the quality and quantity of agricultural yields, as well as affect market prices. This study compares the effectiveness of the YOLO11 and YOLOv8 models in detecting diseases on tomato leaves with traditional CNNbased models such as VGG-16 and Inception-V3. The results show that the YOLO11 model provides the best accuracy of 99.4%, followed by YOLOv8 with 98.5%, both excelling in real-time detection. CNN-based models like VGG-16 and Inception-V3 have high accuracy (99% and 93.8%), but are slower in computation. The ensemble model of VGG-16 and NASNet Mobile achieves an accuracy of 98.7%, but is slightly lower than YOLO11. The YOLO model is more efficient in detection speed, making it a better choice for field applications. This study shows that YOLO11 offers the best combination of accuracy and detection speed for a real-time plant disease detection system.
Sistem Kendali Mikroklimat Greenhouse Cerdas Berbasis Hybrid GA-LSTM dan Fuzzy Logic Controller Rainer Maghma Royali; Dhani Ariatmanto
Jurnal Indonesia : Manajemen Informatika dan Komunikasi Vol. 7 No. 2 (2026): Mei
Publisher : Lembaga Penelitian dan Pengabdian Kepada Masyarakat (LPPM) STMIK Indonesia Banda Aceh

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.63447/jimik.v7i2.1798

Abstract

Controlling temperature and humidity in greenhouses is a complex challenge due to its non-linear nature and dependence on external weather conditions. Conventional control methods often experience energy inefficiency and delayed responses to drastic changes. This study proposes a hybrid approach by combining Long Short-Term Memory (LSTM) optimized using Genetic Algorithm (GA) for temperature prediction, and Fuzzy Logic Controller (FLC) for actuator decision-making. Genetic Algorithm is employed to find optimal hyperparameters (number of neurons and batch size) in the LSTM architecture. Experimental results demonstrate the GA-LSTM model's capability in predicting temperature with high accuracy, yielding an R² score of 0.9881 and Root Mean Square Error (RMSE) of 1.1273°C. These accurate predictions are subsequently used as input to the FLC to regulate exhaust fan speed and mist pump status. Simulations demonstrate the system's capability in making energy-efficient decisions—activating actuators only when conditions are predicted to deviate from ideal values—while remaining responsive to extreme temperature anomalies.
Performance Improvement of CNN Algorithm with Data Augmentation for Tobacco Leaf Disease Classification Rizqy Agung Nurhidayatullah; Ema Utami; Dhani Ariatmanto
Jurnal Teknika Vol 18 No 1 (2026): MARET
Publisher : Universitas Islam Lamongan

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30736/jt.v18i1.1639

Abstract

Pengolahan citra digital daun tembakau merupakan pendekatan penting dalam deteksi dini dan akurat penyakit tanaman, khususnya untuk komoditas pertanian bernilai ekonomi tinggi. Penelitian ini mengusulkan pendekatan augmentasi data multi-strategi untuk meningkatkan akurasi klasifikasi penyakit daun tembakau menggunakan Convolutional Neural Networks (CNN). Tiga teknik augmentasi diterapkan: augmentasi tradisional, augmentasi sampel campuran, dan augmentasi berbasis model pada tiga arsitektur CNN (MobileNetV2, EfficientNetB0, dan VGG16). Dataset terdiri dari 400 citra daun tembakau dengan lima kelas penyakit yang dikumpulkan dari Kabupaten Wonosobo. Eksperimen dilakukan dengan dua skenario pembagian data: 70:30 dan 80:20. Hasil menunjukkan bahwa penerapan augmentasi multi-strategi berhasil meningkatkan akurasi klasifikasi hingga 81,25% pada MobileNetV2 dengan rasio 80:20, yang mewakili peningkatan 20% dibandingkan dengan model tanpa augmentasi. Selain itu, teknik augmentasi secara efektif mengurangi overfitting dengan menurunkan selisih antara akurasi pelatihan dan validasi dari 0,5-0,7 menjadi 0,1-0,2.