Sürdürülebilir Kalkınma Amaçları
Yayınlar
- MakaleAçık Erişim2026Dikkat Tabanlı Görsel–İşitsel Akış Füzyonu ile Sahne Sınıflandırma
Çelik, Yusuf
Bu çalışmada, sahne sınıflandırma görevlerinde görsel ve işitsel verilerin birlikte kullanılmasına yönelik, Gated Cross-Modal Attention Fusion Network (GCM-AFN) adı verilen çok modlu bir derin öğrenme mimarisi önerilmektedir. Önerilen yapıda, görsel akıştan EfficientNetV2 omurgası kullanılarak derin özellikler çıkarılırken, işitsel akışta MFCC tabanlı zaman-frekans temsilleri kullanılmaktadır. Her iki akıştan elde edilen özellik temsilleri, kapılı dikkat mekanizması aracılığıyla birleştirilmektedir. Model, çok kipli bilgilerin etkin biçimde bir araya getirilmesini sağlamak amacıyla yardımcı akış çıktıları ve dikkat tabanlı bir füzyon stratejisi içermektedir. Çapraz doğrulamaya dayalı deneysel sonuçlar, önerilen yaklaşımın tek akışlı modeller ve geleneksel füzyon yöntemlerine kıyasla daha yüksek doğruluk, daha düşük varyans ve sınıf bazında daha tutarlı bir performans sergilediğini göstermektedir. Elde edilen bulgular, dikkat tabanlı füzyonun görsel–işitsel sahne sınıflandırma görevlerinde etkili bir bilgi birleştirme yaklaşımı sunduğunu ortaya koymaktadır.
- MakaleAçık Erişim2026Deep learning-based short-term cryptocurrency price forecasting using LSTM models enhanced with technical indicators
Çelik, Yusuf; İnce, İrem Sevda
This study investigates short-term cryptocurrency price forecasting using deep learning methods. Minute-level price and volume data for BTC/USDT and ETH/USDT trading pairs were obtained through the Binance API and analyzed under different temporal resolutions, including hourly and 15-minute forecasting scenarios. Recurrent neural network architectures, including LSTM, GRU, and BiLSTM, were evaluated using different loss functions to identify effective configurations for short-term forecasting. Experimental results show that deep learning models can achieve high forecasting accuracy for high-frequency cryptocurrency data. In particular, the LSTM model trained with a weighted loss function demonstrated strong and stable prediction performance. In addition, incorporating technical indicators as auxiliary features enriched the feature space and improved forecasting capability. The results also suggest that aligning the temporal resolution of technical indicators with the prediction frequency can further enhance model performance. Overall, the proposed framework provides an effective deep learning–based approach for short-term cryptocurrency price forecasting in high-frequency financial markets.
- MakaleAçık Erişim2026Scopus SJRQ1Deep learning-based detection of bowel sound events in continuous recordings
Çelik, Yusuf
Accurate identification of bowel sound events is essential for the long-term and non-invasive monitoring of gastrointestinal activity. In this study, we present a band-aware multi-band deep learning approach that explicitly accounts for the distinct acoustic characteristics of bowel sounds across different frequency ranges by jointly analyzing low- and high-frequency components. The proposed method operates on time–frequency representations extracted from continuous recordings to produce frame-level bowel sound probabilities, which are subsequently integrated into recording-level decisions through a unified decision-making strategy. Despite the pronounced class imbalance in the dataset, the proposed system achieved accuracy levels exceeding 98% at the frame level and 99% at the recording level, yielding stable and consistent outcomes through probability aggregation and temporal smoothing steps. These results indicate that the proposed approach provides a reliable and well-structured framework for bowel sound event detection and establishes a solid methodological basis for the analysis of long-duration recordings.
AtıfScopus2WoS2OpenAlex2
İncele - MakaleAçık Erişim2026Scopus SJRQ1Surface EMG-Based Hand Gesture Recognition Using a Hybrid Multistream Deep Learning Architecture
Çelik, Yusuf; Can, Umit
Surface electromyography (sEMG) enables non-invasive measurement of muscle activity for applications such as human–machine interaction, rehabilitation, and prosthesis control. However, high noise levels, inter-subject variability, and the complex nature of muscle activation hinder robust gesture classification. This study proposes a multistream hybrid deep-learning architecture for the FORS-EMG dataset to address these challenges. The model integrates Temporal Convolutional Networks (TCN), depthwise separable convolutions, bidirectional Long Short-Term Memory (LSTM)–Gated Recurrent Unit (GRU) layers, and a Transformer encoder to capture complementary temporal and spectral patterns, and an ArcFace-based classifier to enhance class separability. We evaluate the approach under three protocols: subject-wise, random split without augmentation, and random split with augmentation. In the augmented random-split setting, the model attains 96.4% accuracy, surpassing previously reported values. In the subject-wise setting, accuracy is 74%, revealing limited cross-user generalization. The results demonstrate the method’s high performance and highlight the impact of data-partition strategies for real-world sEMG-based gesture recognition.
AtıfScopus4WoS4OpenAlex7İncele - MakaleAçık Erişim2026Scopus SJRQ2A Deep Learning Framework for Local Earthquake Magnitude Estimation Using Three-Component Waveforms
Çelik, Yusuf
This study presents a two-stage deep learning framework for accurate and generalizable estimation of local earthquake magnitudes from three-component seismic waveforms, within the context of ground-based remote sensing systems. In the first stage, phase transition boundaries are identified at the sample level to enable consistent temporal alignment of the signals. In the second stage, earthquake magnitude estimation is performed using 30 s waveform segments aligned with the early portion of the signal and enriched with spectral and statistical features. The model was initially trained on the globally diverse dataset STEAD and later fine-tuned using a subset of KOERI waveforms, and its performance was evaluated on an independent KOERI test set. The results demonstrate high prediction accuracy, with a mean absolute error of approximately 0.09 and a coefficient of determination (R2) of about 0.95, indicating strong agreement between predicted and true magnitudes. The model maintains stable performance across varying signal characteristics and geographic regions, highlighting its strong transferability. These findings suggest that seismic sensor networks can be effectively utilized as remote sensing systems for rapid and reliable earthquake characterization.
- MakaleAçık Erişim2024Application of Deep Learning for Voice Command Classification in Turkish Language
Çelik, Yusuf
In this study, a deep learning model was developed for the recognition and classification of voice commands using the Turkish Speech Command Dataset. The division of training, validation, and test sets was carried out on an individual basis. This approach aims to prevent the model from memorizing and to enhance its generalization capability. The model was trained using Mel-Frequency Cepstral Coefficients (MFCC) features extracted from voice files, and its classification performance was evaluated in detail. The findings indicate that the model successfully classifies voice commands with a high accuracy rate, achieving an overall accuracy of 92.3% on the test set, highlighting the potential of deep learning approaches in voice recognition technologies
AtıfOpenAlex6
İncele - MakaleAçık Erişim2024Bellek Tabanlı LSTM ve GRU Makine Öğrenmesi Algoritmaları Kullanarak BIST100 Endeks Tahmini
Çelik, Yusuf
Makine öğrenmesi tabanlı tahmin yaklaşımlarının finansal piyasalarda geliştirilmesi, hızlı ve hassas karar alma, karmaşıklıkla başa çıkma, risk yönetimi, algoritmik ticaret ve duygusal etkilerin azaltılması gibi avantajlar sağlar. Bu yaklaşımlar, sürekli öğrenme ve adaptasyon yetenekleriyle finansal başarı için rekabet avantajı oluşturabilir. Bu makale çalışmasında, Borsa İstanbul (BIST) 100 endeks tahmini için bellek tabanlı makine öğrenmesi modellerine dayalı bir yaklaşım sunulmuştur. Bu amaçla, ardışık veri değerlendirmesinde popüler olan uzun kısa-süreli bellek (LSTM) ve geçitli tekrarlayan birim (GRU) mimarileri kullanılmıştır. Elde edilen model çıktılarına göre bu modellerin, eğitim ve doğrulama aşamalarında düşük kayıplar gösterdiği ve BIST100 endeksinin genel eğilimlerini başarıyla takip ettiği gözlemlenmiştir. Ancak, modeller piyasa dalgalanmaları ve ani değişimlerde gerçek değerlerden sapmalar göstermiş, bu da belirsizlikleri ve genelleme kapasitelerinin sınırlarını ortaya koymuştur. Geleceğe yönelik tahminler, eğitim veri setindeki desenlere dayanarak yapılmış ancak zamanla artan belirsizlik göstermiştir. Çalışma, makine öğrenmesi algoritmalarının finans verileri üzerindeki kullanım alanı konusunda önemli bilgiler sağlayacak potansiyele sahiptir.
AtıfOpenAlex5İncele - MakaleAçık Erişim2023Scopus SJRQ2Extracting low dimensional representations from large size whole slide images using deep convolutional autoencoders
Çelik, Yusuf; Karabatak, Murat
Background and objective: Artificial intelligence-based analysis of medical images has recently become a trendy field of study. The most significant effect on these systems to produce reliable and high-performance results is the amount of accessed data. Although whole-slide images (WSI), one of the current imaging techniques used in histopathology, storage costs are very expensive. Therefore, low-dimensional visuals representing of WSIs is an essential field of study. Methods: In this study, deep auto-encoder-based models are designed to create low dimensional representations for WSI. The size reduction was performed for the input images in different sizes, at the ratios of 1: 3, 1: 6, and 1: 12, respectively. Results: Similarity index measure values were obtained as high as 0.957 and 0.938, respectively, using 128 × 128 × 3 dimensional patches in the size reduction process of 1: 3 and 1:12 on test WSI. Conclusion: The proposed model has a structure that can be applied in compression of WSIs and secure data transfer. By using these representations, both storage costs and the high-level hardware costs required by deep learning algorithms can be significantly reduced.
AtıfScopus3WoS2OpenAlex4İncele