Feyza Yıldırım Okay
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Feyza Yıldırım Okay has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Suat Ozdemir
Hacettepe University
15 shared publications - Ibrahim Kok
Department of Artificial Intelligence and Data Engineering, Ankara University, 06100 Ankara, Türkiye
7 shared publications - Abdul Amir Alioghli
Department of Computer Science, Gazi University, Ankara, Türkiye
2 shared publications - Ahmet Alkılınç
Department of Computer Engineering, Hacettepe University, Ankara, Turkey
2 shared publications - Oğuzhan Akyıldız2 shared publications
- Özgecan Muyanlı Kopan
Department of Computer Engineering, Hacettepe University, Ankara 06000, Türkiye
2 shared publications
Latest Publications
- Open Access
- 19 October 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (20) , 10347
Are the Eyes a Mirror of Autism? Explainable AI-Based Biomarker Discovery for Diagnosis
Autism spectrum disorder (ASD) presents significant difficulties in diagnosis due to its nature and diversity of symptoms. Traditional diagnostic methods rely on comprehensive clinical evaluations and complex diagnostic processes. Recent advances in machine learning (ML) and explainable artificial intelligence (XAI) offer promising models that help improve diagnostic accuracy and efficiency in ASD. XAI methods provide explainability by determining feature importance and indicating the extent to which features impact the outcome. In this study, we use children’s eye-tracking data and propose using this capability of XAI as a feature selection (FS) method in ASD diagnosis. Specifically, we aim to compare the performance of XAI-based and traditional FS methods individually and explore the benefits of combining these approaches. The goal is to enhance the diagnostic accuracy of ML models while simultaneously identifying significant features—a process referred to as biomarker discovery—through the proposed system. This system enables the identification of critical features that differentiate children with ASD from their typically developing (TD) peers, offering valuable insights to clinicians for improved diagnostic practices. The results demonstrate that the XAI-based FS method improves model performance compared to traditional feature selection techniques, with the highest AUC score of 0.78 achieved by combining feature sets from both methods. Furthermore, the XAI approach successfully identified four candidate biomarkers critical to ASD diagnosis, which were not selected by traditional methods. These findings underscore the potential of XAI techniques not only to boost model accuracy but also to uncover new diagnostic insights. By combining the XAI-based FS method with traditional approaches, this research contributes to advancing diagnostic practices in ASD, identifying new biomarkers, and enhancing clinical decision-making in pediatric neurodevelopmental disorders.
- 19 October 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (20) , 10347
Open AccessAre the Eyes a Mirror of Autism? Explainable AI-Based Biomarker Discovery for Diagnosis
Autism spectrum disorder (ASD) presents significant difficulties in diagnosis due to its nature and diversity of symptoms. Traditional diagnostic methods rely on comprehensive clinical evaluations and complex diagnostic processes. Recent advances in machine learning (ML) and explainable artificial intelligence (XAI) offer promising models that help improve diagnostic accuracy and efficiency in ASD. XAI methods provide explainability by determining feature importance and indicating the extent to which features impact the outcome. In this study, we use children’s eye-tracking data and propose using this capability of XAI as a feature selection (FS) method in ASD diagnosis. Specifically, we aim to compare the performance of XAI-based and traditional FS methods individually and explore the benefits of combining these approaches. The goal is to enhance the diagnostic accuracy of ML models while simultaneously identifying significant features—a process referred to as biomarker discovery—through the proposed system. This system enables the identification of critical features that differentiate children with ASD from their typically developing (TD) peers, offering valuable insights to clinicians for improved diagnostic practices. The results demonstrate that the XAI-based FS method improves model performance compared to traditional feature selection techniques, with the highest AUC score of 0.78 achieved by combining feature sets from both methods. Furthermore, the XAI approach successfully identified four candidate biomarkers critical to ASD diagnosis, which were not selected by traditional methods. These findings underscore the potential of XAI techniques not only to boost model accuracy but also to uncover new diagnostic insights. By combining the XAI-based FS method with traditional approaches, this research contributes to advancing diagnostic practices in ASD, identifying new biomarkers, and enhancing clinical decision-making in pediatric neurodevelopmental disorders.
- 16 September 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 204
BeeCF: Artificial bee colony based counterfactual explanations for healthcare recommendation systems
- 16 September 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 204
BeeCF: Artificial bee colony based counterfactual explanations for healthcare recommendation systems
- Open Access
- 21 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (8) , 1752
FedCycle: An Improved Federated Learning Framework for Assessment Across Modalities and Domains
Artificial Intelligence (AI) systems based on traditional Deep Learning (DL) are expected to play a leading role in the early detection of various diseases in healthcare applications. However, there are two major drawbacks of these systems: protecting patient privacy and obtaining sufficiently large, high-quality datasets to train reliable models. In traditional DL, collecting data from different sources on a single central server increases system complexity and raises serious privacy and security concerns. Federated Learning (FL) makes it possible to train models locally at multiple data locations while collaboratively improving a global model without exposing raw data, making it a promising architectural solution for privacy preservation. Although previous studies have reported that FL can achieve performance comparable to centralized DL approaches, traditional FL approaches often struggle to maintain consistent performance across different settings. This limitation becomes more noticeable when heterogeneous data distributions, modalities, and domains are involved. In these situations, client drift, overfitting, and generalization capability of the global model arise as major challenges. Thus, this study presents FedCycle as an incremental improvement of the FedAvg algorithm. It modifies the aggregation frequency. It aims to overcome these drawbacks and make the global model more stable and efficient. The FedCycle eliminates centralized data collection, enhances data security, and effectively reduces client drift and overfitting by supporting model training across heterogeneous data distributions, modalities, and domains. The performance evaluation involves extensive experiments using various real-world breast cancer image datasets, namely BREAKHIS, ROBOFLOW, RSNA, BUSI, and BCFPP. The presented method is evaluated against both traditional DL and FL approaches using accuracy, precision, recall, F1-score, and AUC. The findings confirm that applying fine-tuning within FedCycle reduces overfitting during training. As a result, FedCycle achieves performance improvements of 7.75% and 4.65% in accuracy and F1-score on the RSNA and BCFPP datasets compared to traditional DL approaches, while also providing an average improvement of approximately 1.5% in accuracy and F1-score across BREAKHIS, ROBOFLOW, and BUSI datasets compared to FedAvg.
- 21 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (8) , 1752
Open AccessFedCycle: An Improved Federated Learning Framework for Assessment Across Modalities and Domains
Artificial Intelligence (AI) systems based on traditional Deep Learning (DL) are expected to play a leading role in the early detection of various diseases in healthcare applications. However, there are two major drawbacks of these systems: protecting patient privacy and obtaining sufficiently large, high-quality datasets to train reliable models. In traditional DL, collecting data from different sources on a single central server increases system complexity and raises serious privacy and security concerns. Federated Learning (FL) makes it possible to train models locally at multiple data locations while collaboratively improving a global model without exposing raw data, making it a promising architectural solution for privacy preservation. Although previous studies have reported that FL can achieve performance comparable to centralized DL approaches, traditional FL approaches often struggle to maintain consistent performance across different settings. This limitation becomes more noticeable when heterogeneous data distributions, modalities, and domains are involved. In these situations, client drift, overfitting, and generalization capability of the global model arise as major challenges. Thus, this study presents FedCycle as an incremental improvement of the FedAvg algorithm. It modifies the aggregation frequency. It aims to overcome these drawbacks and make the global model more stable and efficient. The FedCycle eliminates centralized data collection, enhances data security, and effectively reduces client drift and overfitting by supporting model training across heterogeneous data distributions, modalities, and domains. The performance evaluation involves extensive experiments using various real-world breast cancer image datasets, namely BREAKHIS, ROBOFLOW, RSNA, BUSI, and BCFPP. The presented method is evaluated against both traditional DL and FL approaches using accuracy, precision, recall, F1-score, and AUC. The findings confirm that applying fine-tuning within FedCycle reduces overfitting during training. As a result, FedCycle achieves performance improvements of 7.75% and 4.65% in accuracy and F1-score on the RSNA and BCFPP datasets compared to traditional DL approaches, while also providing an average improvement of approximately 1.5% in accuracy and F1-score across BREAKHIS, ROBOFLOW, and BUSI datasets compared to FedAvg.
- Open Access
- 30 March 2026
- Published by Springer Nature in Cluster Computing
- Vol. 29 (4) , 221
A comprehensive survey on federated semi-supervised learning: applications, challenges, and future directions
Federated Learning (FL) is a decentralized method for collaboratively training a global machine learning model while preserving the privacy of user data. In recent years, its privacy-preserving and collaborative nature has attracted substantial interest among researchers. However, in real-world scenarios, it is often impractical to assume that all clients have fully labeled data, as acquiring large labeled datasets can be costly and requires expertise. To overcome this problem in FL, a new approach called Federated Semi-Supervised Learning (FSSL) is presented which aims to enhance the overall performance of the global model in FL by incorporating unlabeled data during model training. To provide a thorough examination of this area and to stimulate further research, we provide a systematic and comprehensive overview of recent studies using the FSSL method. We also discuss current challenges and open issues to provide researchers with directions for future research.
- 30 March 2026
- Published by Springer Nature in Cluster Computing
- Vol. 29 (4) , 221
Open AccessA comprehensive survey on federated semi-supervised learning: applications, challenges, and future directions
Federated Learning (FL) is a decentralized method for collaboratively training a global machine learning model while preserving the privacy of user data. In recent years, its privacy-preserving and collaborative nature has attracted substantial interest among researchers. However, in real-world scenarios, it is often impractical to assume that all clients have fully labeled data, as acquiring large labeled datasets can be costly and requires expertise. To overcome this problem in FL, a new approach called Federated Semi-Supervised Learning (FSSL) is presented which aims to enhance the overall performance of the global model in FL by incorporating unlabeled data during model training. To provide a thorough examination of this area and to stimulate further research, we provide a systematic and comprehensive overview of recent studies using the FSSL method. We also discuss current challenges and open issues to provide researchers with directions for future research.
- Open Access
- 19 December 2025
- Published by MDPI AG in Sustainability
- Vol. 18 (1) , 24
Deep Ensemble Learning Model for Waste Classification Systems
Waste classification is a critical aspect of sustainable waste management systems. Traditional methods for waste classification are often inadequate to handle the complexity and diversity of materials encountered in real-world scenarios. This paper proposes novel deep ensemble learning models that combine pre-trained models with ensemble methods to improve waste classification performance. The proposed model leverages transfer and ensemble learning techniques, employing both averaging and weighted averaging methods to enhance waste classification accuracy. The proposed model is evaluated comprehensively on four publicly available waste image datasets containing various waste categories: TrashNet, TrashBox, Waste Pictures and Garbage Classification. The obtained results show that the averaging and weighted averaging ensemble methods improved classification accuracy by 1% to 3% over the strongest individual models. The weighted ensemble method achieves 96% accuracy, 94% precision, 97% recall and 95% F1 score on the TrashNet dataset. Statistical significance is verified using 5-fold cross-validation and paired t-tests (p < 0.05). To ensure model explainability, the localization of important object regions is demonstrated with Grad-CAM visualizations. Overall, this study validates the potential of integrating deep image classification models with ensemble methodologies to improve the accuracy and efficiency of waste classification. The main contributions of this study can be summarized as follows: we design an efficient deep ensemble method that leverages multiple pre-trained models and ensemble techniques; we employ averaging and weighted averaging techniques to improve classification accuracy and model robustness; and lastly, we evaluate the model using multiple datasets to demonstrate generalizability, scalability and robustness.
- 19 December 2025
- Published by MDPI AG in Sustainability
- Vol. 18 (1) , 24
Open AccessDeep Ensemble Learning Model for Waste Classification Systems
Waste classification is a critical aspect of sustainable waste management systems. Traditional methods for waste classification are often inadequate to handle the complexity and diversity of materials encountered in real-world scenarios. This paper proposes novel deep ensemble learning models that combine pre-trained models with ensemble methods to improve waste classification performance. The proposed model leverages transfer and ensemble learning techniques, employing both averaging and weighted averaging methods to enhance waste classification accuracy. The proposed model is evaluated comprehensively on four publicly available waste image datasets containing various waste categories: TrashNet, TrashBox, Waste Pictures and Garbage Classification. The obtained results show that the averaging and weighted averaging ensemble methods improved classification accuracy by 1% to 3% over the strongest individual models. The weighted ensemble method achieves 96% accuracy, 94% precision, 97% recall and 95% F1 score on the TrashNet dataset. Statistical significance is verified using 5-fold cross-validation and paired t-tests (p < 0.05). To ensure model explainability, the localization of important object regions is demonstrated with Grad-CAM visualizations. Overall, this study validates the potential of integrating deep image classification models with ensemble methodologies to improve the accuracy and efficiency of waste classification. The main contributions of this study can be summarized as follows: we design an efficient deep ensemble method that leverages multiple pre-trained models and ensemble techniques; we employ averaging and weighted averaging techniques to improve classification accuracy and model robustness; and lastly, we evaluate the model using multiple datasets to demonstrate generalizability, scalability and robustness.
- Open Access
- 26 March 2025
- Published by Gazi University in Gazi University Journal of Science Part A: Engineering and Innovation
- Vol. 12 (1) , 72-95
ChainHealth: Blockchain-Based IoT-Edge Model for Secure Management of Health Data
The Internet of Things (IoT) is rapidly expanding and seamlessly integrating into our daily lives, with an increasing number of objects connecting to the Internet. It operates as a networked architecture that enables communication between connected devices. IoT applications span various domains, including smart homes, cities, transportation, and healthcare. Among these, smart healthcare is particularly important, allowing specialists to monitor patients remotely, anytime, and anywhere. In this system, patient data is transmitted through networked systems, enabling remote health monitoring. However, significant challenges remain regarding the privacy and integrity of patient health data. This study addresses these challenges by proposing a model named ChainHealth that leverages IoT devices for data collection, edge infrastructure for processing, smart contracts on blockchain to ensure data integrity, and blockchain to store data securely. Experimental results demonstrate that ChainHealth significantly outperforms traditional models in terms of data transmission efficiency, scalability, and overall system performance. The model enhances throughput, reduces latency even as the number of users increases, and strengthens data encryption and transmission processes. Additionally, the smart contract mechanism is evaluated and shown to be reliable for managing data integrity. As a result, the proposed model ensures secure data transfer across the network and secure critical health information. By maintaining data integrity, confidentiality, and security, ChainHealth improves both the quality and reliability of healthcare services compared to traditional approaches.
- 26 March 2025
- Published by Gazi University in Gazi University Journal of Science Part A: Engineering and Innovation
- Vol. 12 (1) , 72-95
Open AccessChainHealth: Blockchain-Based IoT-Edge Model for Secure Management of Health Data
The Internet of Things (IoT) is rapidly expanding and seamlessly integrating into our daily lives, with an increasing number of objects connecting to the Internet. It operates as a networked architecture that enables communication between connected devices. IoT applications span various domains, including smart homes, cities, transportation, and healthcare. Among these, smart healthcare is particularly important, allowing specialists to monitor patients remotely, anytime, and anywhere. In this system, patient data is transmitted through networked systems, enabling remote health monitoring. However, significant challenges remain regarding the privacy and integrity of patient health data. This study addresses these challenges by proposing a model named ChainHealth that leverages IoT devices for data collection, edge infrastructure for processing, smart contracts on blockchain to ensure data integrity, and blockchain to store data securely. Experimental results demonstrate that ChainHealth significantly outperforms traditional models in terms of data transmission efficiency, scalability, and overall system performance. The model enhances throughput, reduces latency even as the number of users increases, and strengthens data encryption and transmission processes. Additionally, the smart contract mechanism is evaluated and shown to be reliable for managing data integrity. As a result, the proposed model ensures secure data transfer across the network and secure critical health information. By maintaining data integrity, confidentiality, and security, ChainHealth improves both the quality and reliability of healthcare services compared to traditional approaches.
- Open Access
- 1 March 2025
- Published by Gazi University Journal of Science in Gazi University Journal of Science
- Vol. 38 (1) , 115-129
Fusion of Image Transformation Techniques for IoT-based Multivariate Time-Series
The emergence of the Internet of Things (IoT) has ushered in a new era of data generation with the opportunity for data to become a key element of connected devices. This study investigates new methods to bridge the realms of multivariate time-series data and image analysis, paying special attention to Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP) transformation techniques. These techniques serve to convert raw time-series data into visual representations, laying the foundation for deeper analysis and predictive modeling. The study introduces a novel paradigm by not only employing individual image transformation techniques but also fusing them in both horizontal and square orientations. By leveraging Convolutional Neural Networks (CNNs), this study demonstrates the efficiency of innovative fused-oriented image transformation techniques in predicting complex patterns within a multivariate time-series dataset related to electricity distribution and transformer oil temperature. The experimental results indicate that the Fused-Horizontal image transformation technique, using the order RP - GADF - MTF - GASF, yields the best performance, achieving the lowest MSE of 0.01047, RMSE of 0.10235, and MAE of 0.08054. Additionally, the order RP - GADF - GASF - MTF results in the lowest MAPE of 0.21997, outperforming both Fused-Square techniques and individual methods like GASF, GADF, MTF, and RP. These findings underscore the potential of fused image transformation techniques in improving prediction accuracy, offering a significant advancement over traditional methods.
- 1 March 2025
- Published by Gazi University Journal of Science in Gazi University Journal of Science
- Vol. 38 (1) , 115-129
Open AccessFusion of Image Transformation Techniques for IoT-based Multivariate Time-Series
The emergence of the Internet of Things (IoT) has ushered in a new era of data generation with the opportunity for data to become a key element of connected devices. This study investigates new methods to bridge the realms of multivariate time-series data and image analysis, paying special attention to Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP) transformation techniques. These techniques serve to convert raw time-series data into visual representations, laying the foundation for deeper analysis and predictive modeling. The study introduces a novel paradigm by not only employing individual image transformation techniques but also fusing them in both horizontal and square orientations. By leveraging Convolutional Neural Networks (CNNs), this study demonstrates the efficiency of innovative fused-oriented image transformation techniques in predicting complex patterns within a multivariate time-series dataset related to electricity distribution and transformer oil temperature. The experimental results indicate that the Fused-Horizontal image transformation technique, using the order RP - GADF - MTF - GASF, yields the best performance, achieving the lowest MSE of 0.01047, RMSE of 0.10235, and MAE of 0.08054. Additionally, the order RP - GADF - GASF - MTF results in the lowest MAPE of 0.21997, outperforming both Fused-Square techniques and individual methods like GASF, GADF, MTF, and RP. These findings underscore the potential of fused image transformation techniques in improving prediction accuracy, offering a significant advancement over traditional methods.
- 6 February 2025
- Published by Elsevier in Computer Networks
- Vol. 259
Mx-TORU: Location-aware multi-hop task offloading and resource optimization protocol for connected vehicle networks
- 6 February 2025
- Published by Elsevier in Computer Networks
- Vol. 259
Mx-TORU: Location-aware multi-hop task offloading and resource optimization protocol for connected vehicle networks
- 26 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1) , 1-31
Deepat: a real-time deep learning based model for aircraft tracking system
- 26 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1) , 1-31
Deepat: a real-time deep learning based model for aircraft tracking system
- 11 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1) , 1-27
Enhancing multivariate time-series anomaly detection with positional encoding mechanisms in transformers
- 11 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1) , 1-27
Enhancing multivariate time-series anomaly detection with positional encoding mechanisms in transformers