Claim Missing Document
Check
Articles

Found 34 Documents
Search

Deteksi Indikasi Kelelahan Menggunakan Deep Learning Fudholi, Dhomas Hatta; Nayoan, Royan Abida N; Suyuti, Maghfirah; Rahmadi, Ridho
J-SAKTI (Jurnal Sains Komputer dan Informatika) Vol 5, No 1 (2021): EDISI MARET
Publisher : STIKOM Tunas Bangsa Pematangsiantar

Show Abstract | Download Original | Original Source | Check in Google Scholar | Full PDF (903.898 KB) | DOI: 10.30645/j-sakti.v5i1.292

Abstract

Many students experience fatigue due to lack of sleep which can be caused by a psychological conditions or bad habits. Lack of sleep can affect student’s performance academically and causes many illnesses, stress and depression. Students with fatigue causes students to not study well, increasing risk of academic failure and will lead to having low GPA. In this research, fatigue detection is carried out to find out which students are experiencing fatigue. In this study, an annotated video dataset was used with a total of 18 subjects acted drowsy and alert. Fatigue detection is based on mouth movements, therefore mouth annotation is used. Mouth annotation has 2 categories, namely annotation 0 which indicates a closed mouth and annotation 1 which indicates the mouth is yawning. Previous study proves ResNet50 has better performance than other pre-trained models such as AlexNet, Clarifia, VGG-16, and GoogLeNet-19. We also applied image augmentation which is useful for providing new image variations to the model in each epoch by changing the rotation, random shift, and random zoom. ResNet50 model is used to perform binary classification which has two outputs, namely mouth stillness and yawning. The results of the frame classification are evaluated using precision, recall and f1-score. By using ResNet model, the results of the classification of frames labeled 0 or mouth stillness, obtained a precision of 0.72, a recall of 0.88, and an f1-score of 0.79. Meanwhile, the frame classification labeled 1 or yawning has a precision value of 0.85, a recall of 0.65, and an f1-score of 0.74.
Deteksi Cyberbullying berdasarkan Unsur Perbuatan Pidana yang Dilanggar dengan Naive Bayes dan Support Vector Machine Manoppo, Tommy Nugraha; Fudholi, Dhomas Hatta
J-SAKTI (Jurnal Sains Komputer dan Informatika) Vol 5, No 1 (2021): EDISI MARET
Publisher : STIKOM Tunas Bangsa Pematangsiantar

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30645/j-sakti.v5i1.293

Abstract

Lack of understanding by Indonesian social media user about law impact inflicted to cyberbullying perpetrators makes many cyberbullying cases has not handled properly and ended up with nothing. Indonesia hasn’t yet law authority that govern cyberbullying in specific, causing no guideline regard the definition about cyberbullying itself. There is an extension about definition of violence which state that violence is not only physically deliver, but also psychologically, referred an inferences cyberbullying characteristics possibly qualify in element of criminal act. Therefore, the element of criminal act can be used as a basis for detecting potential of cyberbullying. In this research, literature review is used to determine the elements of criminal acts related to the characteristics of cyberbullying and also in finding a model classifier to detect cyberbullying messages. So there are 5 criminal acts related to cyberbullying characteristic which insult, accuse with defamation, hatred about ethnicity, religion, race and inter-group relations, threat of violence, and threat of telling secret. Total of 5000 tweets are collected as a dataset. Feature extraction, using the N-gram method with TF-IDF weighting is expected to obtain sentiment based on the use of words. The context of language becomes important in this study, so the dataset annotation process is carried out by linguist. The results on the application of the two model classfier were Naïve Bayes and SVM after applying resampling by over-sampling using SMOTE method, can correctly predict the potential for cyberbullying by their violated element of criminal act with the average performance measurement of 90%.
Model Identifikasi Penyakit Pada Tumbuhan Padi Berbasiskan DenseNet Pailus, Muhammad; Fudholi, Dhomas Hatta; Hidayat, Syarif
J-SAKTI (Jurnal Sains Komputer dan Informatika) Vol 6, No 2 (2022): EDISI SEPTEMBER
Publisher : STIKOM Tunas Bangsa Pematangsiantar

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30645/j-sakti.v6i2.478

Abstract

Errors in identifying diseases in rice plants can cause the potential for crop failure to increase by 18-80%, according to data from the Indonesian Ministry of Agriculture. This could be due to the lack of expertise in agriculture when compared to the amount of land in Indonesia. Recent research in the field of deep learning using neural networks has achieved remarkable improvements. Research on the identification of plant diseases in rice plants, using the MobileNet, NasNet and SqueezeNet architecture that supports mobile devices has been carried out. The experimental results show that the proposed architecture can achieve an accuracy of 93.3%. Motivated by previous research, this research will use DenseNet architecture (Dense Convolutional Network) to detect diseases in rice plants. The dataset used is relatively small, between 100-200 photos for each disease. To cover the lack of dataset augmentation is done to the dataset. The final results obtained are quite satisfactory with an accuracy of 96% with a Weighted Average of 97%.
Story Generator Bahasa Indonesia dengan Skip-Thoughts Mustofa, M; Fudholi, Dhomas Hatta
J-SAKTI (Jurnal Sains Komputer dan Informatika) Vol 6, No 2 (2022): EDISI SEPTEMBER
Publisher : STIKOM Tunas Bangsa Pematangsiantar

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30645/j-sakti.v6i2.479

Abstract

Currently, there are many studies that want computers to be able to imitate human creativity in stringing words into writing like a writer. This study aims to use the RNN algorithm to produce automatic story writing in Indonesian. The main contribution in this research is the creation and evaluation of the RNN algorithm based on the skip-thoughts model using an Indonesian language dataset. The skip-thoughts model consists of an encoder in the form of single GRU layer with 500 hidden units, and two decoders with single GRU layer each with 500 hidden units. The function of the encoder is to do the word mapping process from the input sentence, while the decoder predicts the sentence before (previous decoder) and the sentence after (next decoder) from the input sentence. The dataset used in the model training is in the form of stories in Indonesian with the genres of folklore and short stories. The model training process is run in 100 epochs, using the ADAM optimizer to get the optimal model. Based on the results of the assessment of respondents who have a background as writers, the folklore model shows a fairly good rating (average score of 65) for the S-P-O-K criteria, and a low rating for criteria of linkage between sentences (average score of 38) and the context of the whole story (average score of 32). The short story of life model shows a good rating (average score of 73) for the S-P-O-K criteria, and a low rating for the linkage between sentences criteria (average score of 48), and the context of the whole story (average score of 42). Based on the results of the assessment, the skip-thoughts model used in the Indonesian story generator has worked well, but it can still be improved by increasing the number of training datasets for each story genre used, as well as being more specific in determining the genre in order to obtain story integrity better.
YOLO-based Small-scaled Model for On-Shelf Availability in Retail Fudholi, Dhomas Hatta; Kurniawardhani, Arrie; Andaru, Gabriel Imam; Alhanafi, Ahmad Azzam; Najmudin, Nabil
Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi) Vol 8 No 2 (2024): April 2024
Publisher : Ikatan Ahli Informatika Indonesia (IAII)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29207/resti.v8i2.5600

Abstract

The availability of the shelf (OSA) in the retail industry plays a very crucial role in continuous sales. Unavailability of products can make a bad impression on customers and reduce sales. The retail industry may continue to develop through the rapidly advancing technology era to thrive in a market where competition is increasingly tough. Along with technological advances in recent decades, artificial intelligence has begun to be applied to support OSA, particularly by using object detection technology. In this research, we develop a small-scale object detection model based on four versions of the You Only Look Once (YOLO) algorithm, namely YOLOv5-nano, YOLOv6-nano, YOLOv7-tiny, and YOLOv8-nano. The developed model can be used to support automatic detection of OSA. A small-scale model has developed in the sense of postpractical implementation through low-cost mobile applications. We also use the quantization method to reduce the model size, INT8 and FP16. This small-scale model implementation also offers flexibility in implementation. With a total of 7697 milk-based retail product images and 125 different product classes, the experiment results show that the developed YOLOv8-nano model, with a mAP50 score of 0.933 and an inference time of 13.4 ms, achieved the best performance.
Lightweight Models for Real-Time Steganalysis: A Comparison of MobileNet, ShuffleNet, and EfficientNet Bauravindah, Achmad; Fudholi, Dhomas Hatta
Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi) Vol 8 No 6 (2024): December 2024
Publisher : Ikatan Ahli Informatika Indonesia (IAII)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29207/resti.v8i6.6091

Abstract

In the digital age, the security of communication technologies is paramount, with cybercrime projected to reach $10.5 trillion annually by 2025. While encryption is vital, decrypted data remains vulnerable, prompting the exploration of steganography as an additional security layer. Steganography conceals data within digital media, but its misuse for cyberattacks—such as embedding malware—has highlighted the need for steganalysis, the detection of hidden data. Despite extensive research, few studies have explored lightweight deep learning models for real-time steganalysis in resource-constrained environments like mobile devices. This research evaluates MobileNet, ShuffleNet, and EfficientNet for such tasks, using the BOSSbase-1.01 dataset. Models were assessed based on accuracy, computational efficiency, and resource usage. MobileNet achieved the highest computational speed but with only 63.8% accuracy, falling short of practical application. ShuffleNet and EfficientNet performed at random-guessing levels with 50% accuracy, reflecting the challenges of steganalysis on mobile platforms. Future work aims to improve accuracy by integrating advanced preprocessing techniques, attention mechanisms, and hybrid architectures, as well as leveraging ensemble methods for improved detection. Data augmentation, transfer learning, and hyperparameter tuning will also be explored to optimize model performance. This study contributes by identifying these challenges and offering insights for future research, focusing on optimizing models and preprocessing techniques to enhance detection accuracy in resource-constrained environments.
ENHANCING CUSTOMER INSIGHT THROUGH ASPECT-BASED SENTIMENT ANALYSIS OF SMART DEVICE REVIEWS : CROSS-BRAND INSIGHTS FROM APPLE, SAMSUNG, AND XIAOMI Al-Sabahi, Abdullah Yahya Moqbel; Fudholi, Dhomas Hatta
PENDIDIKAN SAINS DAN TEKNOLOGI Vol 12 No 3 (2025)
Publisher : STKIP PGRI Situbondo

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47668/edusaintek.v12i3.1796

Abstract

In today’s highly competitive smart device market, understanding customer sentiment is essential for driving product innovation and brand loyalty. This study applies Aspect-Based Sentiment Analysis (ABSA) using transformer-based models BERT, RoBERTa, and DistilBERT on over 20,000 reviews collected from Amazon, Flipkart, Walmart, and Best Buy between 2021 and 2024. The analysis focuses on customer feedback regarding Apple, Samsung, and Xiaomi smartphones and smartwatches. Key product aspects such as Battery, Camera, and Performance were extracted, and sentiment trends were compared by brand and device type. Findings reveal Samsung devices received the highest engagement, with its watches praised but phones criticized. Xiaomi’s reviews showed strong polarization, while Apple maintained consistent but lower review volumes. Temporal trends showed a significant rise in positive sentiment in 2024, indicating improving product satisfaction. This research offers actionable insights for original equipment manufacturers (OEMs) and marketers, highlighting which features drive satisfaction and how sentiment evolves over time.
Pengembangan Chatbot Informasi Hukum Layanan Publik Berbasis Retrieval-Augmented Generation Menggunakan LangChain dan OpenAI di Ombudsman DIY Yasmin, Saarah Muthiah; Fudholi, Dhomas Hatta
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.995

Abstract

Hubungan antara masyarakat dan lembaga layanan publik kerap menghadapi berbagai tantangan, khususnya dalam praktik maladministrasi, penyalahgunaan wewenang, dan kurangnya transparansi. Untuk menjawab permasalahan tersebut, penelitian ini mengusulkan pengembangan chatbot berbasis kecerdasan buatan generatif dengan pendekatan Retrieval-Augmented Generation (RAG) menggunakan LangChain dan OpenAI. Sistem ini dirancang untuk menyajikan informasi hukum yang akurat, kontekstual, dan mudah dipahami oleh masyarakat. Metode yang digunakan dalam penelitian ini meliputi perancangan sistem chatbot dengan arsitektur RAG menggunakan LangChain dan OpenAI. Dokumen hukum diolah menjadi embeddings, disimpan dalam basis data vektor Chroma, dan digunakan dalam proses prompt engineering untuk menghasilkan jawaban yang kontekstual. Evaluasi sistem dilakukan melalui penyebaran kuesioner kepada lima ahli dari Lembaga Ombudsman DIY, dengan analisis data menggunakan pendekatan deskriptif. Evaluasi sistem dilakukan dengan memberikan kuesioner kepada lima ahli dari LO DIY dan menggunakan analisis deskrpitif. Hasil evaluasi menunjukkan bahwa sistem memperoleh skor tinggi dalam indikator Perceived Usefulness (rata – rata = 12) dan  Relevansi (rata – rata = 8), serta skor sangat tinggi dalam indikator Akurasi (rata – rata = 18,6)  dan indikator Clarity (rata – rata = 8,4). Dengan demikian, penerapan teknologi RAG dalam pengembangan chatbot berpotensi meningkatkan pemahaman masyarakat terhadap hukum layanan publik serta memperkuat transparansi dan akuntabilitas dalam penyelenggaraan pelayanan publik. Hal ini menunjukkan potensi strategis pemanfaatan AI dalam mendorong tata kelola pelayanan publik yang lebih responsif, akuntabel, dan inklusif.
Efficient Thoracic Abnormalities Detection Using Mobile Deep Learning Models Bauravindah, Achmad; Fudholi, Dhomas Hatta; Wahyuningrum, Rima Tri
Kinetik: Game Technology, Information System, Computer Network, Computing, Electronics, and Control Vol. 10, No. 3, August 2025
Publisher : Universitas Muhammadiyah Malang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.22219/kinetik.v10i3.2268

Abstract

Indonesia faces a critical shortage of radiologists, with only 1.2 radiologists per 100,000 individuals. This shortage leads to delays in diagnosing thoracic abnormalities such as pneumothorax, cardiomegaly, nodule/mass, consolidation, and infiltration. Chest X-ray (CXR) interpretation remains challenging due to overlapping radiological features, necessitating AI-assisted solutions. This study evaluates three lightweight deep learning models—MobileNetV2, ShuffleNetV2, and EfficientNetB0—for automated thoracic abnormality detection using the ChestX-ray8 dataset. We assessed model performance using accuracy, precision, recall, F1-score, and AUC-ROC, selecting the best model based on the highest per-fold F1-score. EfficientNetB0 emerged as the top-performing model, achieving a macro-average F1-score of 0.556 and AUC-ROC of 0.765, outperforming MobileNetV2 (0.494, 0.719) and ShuffleNetV2 (0.481, 0.713). Grad-CAM analysis revealed strong localization for pneumothorax and consolidation but misclassifications in cardiomegaly and nodule/mass detection due to poor feature differentiation. The findings highlight EfficientNetB0’s potential as an AI-assisted diagnostic tool for low-resource settings while also underscoring the need for segmentation-based pretraining and multi-scale feature extraction to enhance detection accuracy. Future work should focus on optimizing sensitivity to subtle abnormalities and ensuring clinical trust through improved interpretability techniques.
Mi-Botway: a Deep Learning-based Intelligent University Enquiries Chatbot Windiatmoko, Yurio; Hidayatullah, Ahmad Fathan; Fudholi, Dhomas Hatta; Rahmadi, Ridho
International Journal of Artificial Intelligence Research Vol 6, No 1 (2022): June 2022
Publisher : Universitas Dharma Wacana

Show Abstract | Download Original | Original Source | Check in Google Scholar | Full PDF (478.614 KB) | DOI: 10.29099/ijair.v6i1.247

Abstract

Intelligent systems for universities that are powered by artificial intelligence have been developed on a large scale to help people with various tasks. The chatbot concept is nothing new in today's society, which is developing with the latest technology. Students or prospective students often need actual information, such as asking customer service about the university, especially during the current pandemic, when it is difficult to hold a personal meeting in person. Chatbots utilized functionally as lecture schedule information, student grades information, also with some additional features for Muslim prayer schedules and weather forecast information. This conversation bot was developed with a deep learning model adopted by an artificial intelligence model that replicates human intelligence with a specific training scheme. The deep learning implemented is based on RNN which has a special memory storage scheme for deep learning models, in particular in this conversation bot using GRU which is integrated into RASA chatbot framework. GRU is also known as Gated Recurrent Unit, which effectively stores a portion of the memory that is needed, but removes the part that is not necessary. This chatbot is represented by a web application platform created by React JavaScript, and has 0.99 Average Precision Score.