Özgür Can Turna
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Özgür Can Turna has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Muhammed Ali Aydin
İstanbul Üniversitesi-Cerrahpaşa
8 shared publications H. Zaim
4 shared publicationsT. Atmaca
4 shared publications- Zeynep Gurkas Aydin
Department Of Computer Engineering, Istanbul University-Cerrahpasa, Istanbul, Turkiye
3 shared publications A. Halim Zaim
Institute of Science and Engineering, Istanbul Commerce University, Istanbul, Turkey
2 shared publications- Mehmet Ali Erturk
School of Computing, Engineering and Built Environment, Edinburgh Napier University, Edinburgh, U.K.
2 shared publications
Latest Publications
- 15 July 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (4) , 2039-2054
Privacy-Aware Adaptive Differential Privacy for Semantic Retrieval: A Pii-Aware Dynamic Budget Allocation Framework
Retrieval-Augmented Generation (RAG) makes it possible to enhance output from Large Language Models (LLMs) with retrieval-based evidence; however, these systems can also introduce privacy vulnerabilities, and in particular, Membership Inference Attacks (MIAs). Specifically, Differential Privacy (DP) is a well-established methodology for protecting privacy; however, applying static perturbation strategies, typically used in low-dimensional spaces, will result in degradation of semantic utility due to the high dimensionality of the embedding spaces. To solve this privacy-utility problem, we present PADP, a sensitivity-aware perturbation framework inspired by differential privacy principles, which provides a plug-and-play, middleware framework that can be easily integrated into enterprise RAG pipelines without requiring costly computations for LLM fine-tuning and reconstruction of vector indices. PADP uses Named Entity Recognition (NER) based on DeBERTa-v3 to perform sensitivity assessments of documents at the document level on an offline basis, computing PII sensitivity scores for each document. These scores are then used to adaptively calibrate the application of Laplace perturbations to dense query embeddings during retrieval. PADP assigns more protection to the higher-risk content and less to the lower-risk content than static perturbation methods would typically do, thereby increasing the effectiveness of retrieval. PADP draws on concepts from Differential Privacy; nonetheless, it does not furnish formal database-level Differential Privacy guarantees under conventional adjacency definitions. Consequently, the proposed framework should be understood as an adaptive privacy-enhancement mechanism rather than as a formally guaranteed Differential Privacy mechanism. Through the evaluation of the performance of PADP across diverse datasets by way of retrieval effectiveness metrics and distinguishability analysis, its comparison with static embedding perturbation baseline methods affords a contextual statistical analysis which indicates that adaptive sensitivity aware perturbation preserves semantic retrievability while dampening membership based distinguishability signals under the conditions of the study. These findings position sensitivity-aware perturbation as a practical approach to enhancing privacy in Retrieval-Augmented Generation systems while preserving utility.
- 15 July 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (4) , 2039-2054
Privacy-Aware Adaptive Differential Privacy for Semantic Retrieval: A Pii-Aware Dynamic Budget Allocation Framework
Retrieval-Augmented Generation (RAG) makes it possible to enhance output from Large Language Models (LLMs) with retrieval-based evidence; however, these systems can also introduce privacy vulnerabilities, and in particular, Membership Inference Attacks (MIAs). Specifically, Differential Privacy (DP) is a well-established methodology for protecting privacy; however, applying static perturbation strategies, typically used in low-dimensional spaces, will result in degradation of semantic utility due to the high dimensionality of the embedding spaces. To solve this privacy-utility problem, we present PADP, a sensitivity-aware perturbation framework inspired by differential privacy principles, which provides a plug-and-play, middleware framework that can be easily integrated into enterprise RAG pipelines without requiring costly computations for LLM fine-tuning and reconstruction of vector indices. PADP uses Named Entity Recognition (NER) based on DeBERTa-v3 to perform sensitivity assessments of documents at the document level on an offline basis, computing PII sensitivity scores for each document. These scores are then used to adaptively calibrate the application of Laplace perturbations to dense query embeddings during retrieval. PADP assigns more protection to the higher-risk content and less to the lower-risk content than static perturbation methods would typically do, thereby increasing the effectiveness of retrieval. PADP draws on concepts from Differential Privacy; nonetheless, it does not furnish formal database-level Differential Privacy guarantees under conventional adjacency definitions. Consequently, the proposed framework should be understood as an adaptive privacy-enhancement mechanism rather than as a formally guaranteed Differential Privacy mechanism. Through the evaluation of the performance of PADP across diverse datasets by way of retrieval effectiveness metrics and distinguishability analysis, its comparison with static embedding perturbation baseline methods affords a contextual statistical analysis which indicates that adaptive sensitivity aware perturbation preserves semantic retrievability while dampening membership based distinguishability signals under the conditions of the study. These findings position sensitivity-aware perturbation as a practical approach to enhancing privacy in Retrieval-Augmented Generation systems while preserving utility.
- 1 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Bandwidth allocation and energy efficiency solutions in optical access networks
Allocation des ressources et des solutions pour économiser de l'énergie dans les réseaux optiques d'accès: Dans ce travail de thèse, un aperçu général sur les systèmes PON est présenté et sont étudiés les mécanismes et classification PON existants. Après, nous introduisons notre première contribution qui est un algorithme d'allocation dynamique de bande passante pour EPON. Cet algorithme proposé est désigné comme «hcDBA». Par la suite, une amélioration de l'algorithme de hcDBA avec mécanisme de prédiction précoce est présentée. Notre simulation montre bien que notre algorithme hcDBA est performant comparé aux mécanismes existants en termes de taux de perte de paquets et de délai d’accès. Dans notre seconde contribution, nous sommes intéressés au problème de consommation d’énergie qui est un sujet d’actualité dans les réseaux de télécommunication. Les études montrent aujourd’hui que les réseaux d'accès constituent une partie remarquable de la consommation totale d'énergie dans les réseaux de télécommunication. Les organisations ITU-T et IEEE ont publié la recommandation pour la conversation de l'énergie pour les réseaux PONs. Bien que, la consommation totale d'énergie des nœuds ONU est plus importantes que d'autres équipements dans le réseau d'accès fixe, les normes et la plupart des travaux de recherches ont porté sur les économies d'énergie du côté de ONU. Dans cette thèse, nous sommes concentrés sur une méthode d'efficacité énergétique basée sur la conservation de l'énergie du côté de l’OLT. La méthode proposée permet d’économiser de l'énergie en déplaçant dynamiquement des cartes d’OLT en mode de sommeil profond en fonction des charges de trafic entrant et sortant
- 1 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Bandwidth allocation and energy efficiency solutions in optical access networks
Allocation des ressources et des solutions pour économiser de l'énergie dans les réseaux optiques d'accès: Dans ce travail de thèse, un aperçu général sur les systèmes PON est présenté et sont étudiés les mécanismes et classification PON existants. Après, nous introduisons notre première contribution qui est un algorithme d'allocation dynamique de bande passante pour EPON. Cet algorithme proposé est désigné comme «hcDBA». Par la suite, une amélioration de l'algorithme de hcDBA avec mécanisme de prédiction précoce est présentée. Notre simulation montre bien que notre algorithme hcDBA est performant comparé aux mécanismes existants en termes de taux de perte de paquets et de délai d’accès. Dans notre seconde contribution, nous sommes intéressés au problème de consommation d’énergie qui est un sujet d’actualité dans les réseaux de télécommunication. Les études montrent aujourd’hui que les réseaux d'accès constituent une partie remarquable de la consommation totale d'énergie dans les réseaux de télécommunication. Les organisations ITU-T et IEEE ont publié la recommandation pour la conversation de l'énergie pour les réseaux PONs. Bien que, la consommation totale d'énergie des nœuds ONU est plus importantes que d'autres équipements dans le réseau d'accès fixe, les normes et la plupart des travaux de recherches ont porté sur les économies d'énergie du côté de ONU. Dans cette thèse, nous sommes concentrés sur une méthode d'efficacité énergétique basée sur la conservation de l'énergie du côté de l’OLT. La méthode proposée permet d’économiser de l'énergie en déplaçant dynamiquement des cartes d’OLT en mode de sommeil profond en fonction des charges de trafic entrant et sortant
- Top Cited PapersOpen Access
Performance Analysis of LSTM, GRU and Hybrid LSTM–GRU Model for Detecting GPS Spoofing Attacks
The exposure of Unmanned Aerial Vehicles (UAVs) to Global Positioning System (GPS) spoofing attacks constitutes a major cybersecurity challenge. In this work, we conduct a comparative performance analysis of LSTM, GRU, and sequential LSTM–GRU hybrid deep learning models for the detection of GPS spoofing attacks. The ‘UAV Attack’ dataset was preprocessed, and the 11 most significant features were selected using correlation and mutual information algorithms. The models were evaluated using a robust 5-fold cross-validation framework. A combination of 99.31% accuracy, 96.98% recall, and a 97.47% F1-score was achieved by the LSTM–GRU hybrid model, distinguishing it as the leading performer in the experimental study. The LSTM model achieved the highest precision, with a value of 98.49%. ROC curves and AUC values confirmed that the classification performance of all models was close to perfect for the simulated dataset. The findings indicate that deep-learning-based models incorporating the hybrid LSTM–GRU architectures provide an effective and reliable approach designed to identify GPS-spoofing threats affecting UAVs.
Top Cited PapersOpen AccessPerformance Analysis of LSTM, GRU and Hybrid LSTM–GRU Model for Detecting GPS Spoofing Attacks
The exposure of Unmanned Aerial Vehicles (UAVs) to Global Positioning System (GPS) spoofing attacks constitutes a major cybersecurity challenge. In this work, we conduct a comparative performance analysis of LSTM, GRU, and sequential LSTM–GRU hybrid deep learning models for the detection of GPS spoofing attacks. The ‘UAV Attack’ dataset was preprocessed, and the 11 most significant features were selected using correlation and mutual information algorithms. The models were evaluated using a robust 5-fold cross-validation framework. A combination of 99.31% accuracy, 96.98% recall, and a 97.47% F1-score was achieved by the LSTM–GRU hybrid model, distinguishing it as the leading performer in the experimental study. The LSTM model achieved the highest precision, with a value of 98.49%. ROC curves and AUC values confirmed that the classification performance of all models was close to perfect for the simulated dataset. The findings indicate that deep-learning-based models incorporating the hybrid LSTM–GRU architectures provide an effective and reliable approach designed to identify GPS-spoofing threats affecting UAVs.
- Open Access
- 3 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (1) , 488
From Data to Autonomy: Integrating Demographic Factors and AI Models for Expert-Free Exercise Coaching
This study investigates the performance of three deep learning architectures—LSTM with Attention, GRU with Attention, and Transformer—in the context of real-time, self-guided exercise classification, using coordinate data collected from 103 participants via a dual-camera system. Each model was evaluated over ten randomized runs to ensure robustness and statistical validity. The GRU + Attention and LSTM + Attention models demonstrated consistently high test accuracy (mean ≈ 98.9%), while the Transformer model yielded significantly lower accuracy (mean ≈ 96.6%) with greater variance. Paired t-tests confirmed that the difference between LSTM and GRU models was not statistically significant (p = 0.9249), while both models significantly outperformed the Transformer architecture (p < 0.01). In addition, participant-specific features, such as athletic experience and BMI, were found to affect classification accuracy. These findings support the feasibility of AI-based feedback systems in enhancing unsupervised training, offering a scalable solution to bridge the gap between expert supervision and autonomous physical practice.
- 3 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (1) , 488
Open AccessFrom Data to Autonomy: Integrating Demographic Factors and AI Models for Expert-Free Exercise Coaching
This study investigates the performance of three deep learning architectures—LSTM with Attention, GRU with Attention, and Transformer—in the context of real-time, self-guided exercise classification, using coordinate data collected from 103 participants via a dual-camera system. Each model was evaluated over ten randomized runs to ensure robustness and statistical validity. The GRU + Attention and LSTM + Attention models demonstrated consistently high test accuracy (mean ≈ 98.9%), while the Transformer model yielded significantly lower accuracy (mean ≈ 96.6%) with greater variance. Paired t-tests confirmed that the difference between LSTM and GRU models was not statistically significant (p = 0.9249), while both models significantly outperformed the Transformer architecture (p < 0.01). In addition, participant-specific features, such as athletic experience and BMI, were found to affect classification accuracy. These findings support the feasibility of AI-based feedback systems in enhancing unsupervised training, offering a scalable solution to bridge the gap between expert supervision and autonomous physical practice.
- Open Access
- 18 December 2025
- Published by Istanbul Ticaret Universitesi in İstanbul Ticaret Üniversitesi Fen Bilimleri Dergisi
- Vol. 24 (48) , 465-476
SUSTAINABLE ENERGY SOLUTIONS FOR MOBILE NETWORK INFRASTRUCTURE
This article examines the importance and application possibilities of sustainable energy solutions in mobile network infrastructures. Reducing dependency on traditional energy sources provides environmental sustainability in the mobile communication sector while also providing long-term cost savings. The article discusses the integration of renewable energy sources, technologies that increase energy efficiency, and technological innovations in this field. In addition, economic and environmental benefits, challenges, and strategies to overcome these challenges are discussed. Finally, future research areas and application opportunities are evaluated, emphasizing the importance of sustainable solutions in energy management of mobile network infrastructures.
- 18 December 2025
- Published by Istanbul Ticaret Universitesi in İstanbul Ticaret Üniversitesi Fen Bilimleri Dergisi
- Vol. 24 (48) , 465-476
Open AccessSUSTAINABLE ENERGY SOLUTIONS FOR MOBILE NETWORK INFRASTRUCTURE
This article examines the importance and application possibilities of sustainable energy solutions in mobile network infrastructures. Reducing dependency on traditional energy sources provides environmental sustainability in the mobile communication sector while also providing long-term cost savings. The article discusses the integration of renewable energy sources, technologies that increase energy efficiency, and technological innovations in this field. In addition, economic and environmental benefits, challenges, and strategies to overcome these challenges are discussed. Finally, future research areas and application opportunities are evaluated, emphasizing the importance of sustainable solutions in energy management of mobile network infrastructures.
- 30 September 2025
- Published by Academic Platform Journal of Engineering and Smart Systems in Academic Platform Journal of Engineering and Smart Systems
- Vol. 13 (3) , 94-107
Advancements in Human Pose Estimation: A Review of Key Studies and Findings till 2024
This paper presents an in-depth literature review that comprehensively covers the major developments, methods, architectures and datasets used in the field of human pose prediction up to 2025. The review covers a broad spectrum, starting with traditional methods, deep learning-based techniques, convolutional neural networks, graph-based approaches and more recently prominent transformer-based models. In addition to two-dimensional (2D) and three-dimensional (3D) human pose estimation methods, the paper analyses in detail the diversity of data sets, applications of Microsoft Kinect technology, real-time pose estimation systems and related architectural designs. Overall, the review of more than 120 papers shows that existing systems have made significant progress in terms of accuracy, computational efficiency and practical applications, but that there are still some challenges to overcome in complex scenarios such as multiple person detection, occlusion problems and outdoor environments. This in-depth analysis highlights current trends in the field, future research directions and potential applications.
- 30 September 2025
- Published by Academic Platform Journal of Engineering and Smart Systems in Academic Platform Journal of Engineering and Smart Systems
- Vol. 13 (3) , 94-107
Advancements in Human Pose Estimation: A Review of Key Studies and Findings till 2024
This paper presents an in-depth literature review that comprehensively covers the major developments, methods, architectures and datasets used in the field of human pose prediction up to 2025. The review covers a broad spectrum, starting with traditional methods, deep learning-based techniques, convolutional neural networks, graph-based approaches and more recently prominent transformer-based models. In addition to two-dimensional (2D) and three-dimensional (3D) human pose estimation methods, the paper analyses in detail the diversity of data sets, applications of Microsoft Kinect technology, real-time pose estimation systems and related architectural designs. Overall, the review of more than 120 papers shows that existing systems have made significant progress in terms of accuracy, computational efficiency and practical applications, but that there are still some challenges to overcome in complex scenarios such as multiple person detection, occlusion problems and outdoor environments. This in-depth analysis highlights current trends in the field, future research directions and potential applications.
- Open Access
- 10 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11145
A Hybrid Genetic Algorithm-Based Random Forest Model for Intrusion Detection Approach in Internet of Medical Things
The Internet of Medical Things (IoMT) is a bio-network of associated medical devices, which is slowly improving the healthcare industry by focusing its abilities on enhancing personal healthcare benefits with medical data. Moreover, the IoMT tries to deliver sufficient and more suitable medical services at a low cost. With the rapid growth of technology, medical instruments that are widely used anywhere are likely to increase security issues and create safe data transmission issues through resource limitations and available connectivity. Moreover, the patients probably face the risk of different forms of physical harm because of IoMT device attacks. In this paper, we present a secure environment for IoMT devices against cyber-attacks for patient medical data using a new IoMT framework with a hybrid genetic algorithm-based random forest (GA-RF) model. The proposed algorithm achieved better results in terms of accuracy (99.999%), precision, and recall (100%, respectively) to detect cyber-attacks based on two NSL-KDD and UNSW_2018_IoT_Botnet data sets than the other machine learning algorithms.
- 10 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11145
Open AccessA Hybrid Genetic Algorithm-Based Random Forest Model for Intrusion Detection Approach in Internet of Medical Things
The Internet of Medical Things (IoMT) is a bio-network of associated medical devices, which is slowly improving the healthcare industry by focusing its abilities on enhancing personal healthcare benefits with medical data. Moreover, the IoMT tries to deliver sufficient and more suitable medical services at a low cost. With the rapid growth of technology, medical instruments that are widely used anywhere are likely to increase security issues and create safe data transmission issues through resource limitations and available connectivity. Moreover, the patients probably face the risk of different forms of physical harm because of IoMT device attacks. In this paper, we present a secure environment for IoMT devices against cyber-attacks for patient medical data using a new IoMT framework with a hybrid genetic algorithm-based random forest (GA-RF) model. The proposed algorithm achieved better results in terms of accuracy (99.999%), precision, and recall (100%, respectively) to detect cyber-attacks based on two NSL-KDD and UNSW_2018_IoT_Botnet data sets than the other machine learning algorithms.
- 8 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (2) , 2776-2783
ICON: Instagram Profile Classification Using Image and Natural Language Processing Methods
The use of social media has grown significantly, and businesses are now using these platforms to promote their products and services. To do this, companies have created business accounts on social networks. However, social media platforms can also be a breeding ground for unwanted behaviors such as cyberbullying, sexual content, and promotional comments. To address this issue, a study was conducted to create a system that could classify public accounts on Instagram by analyzing comments, profile pictures, bios, and posts shared by users with business accounts. First, a crawler was developed, and data were collected using this crawler and then anonymized. Next, the collected data were processed using natural language processing (NLP) techniques for text and image processing methods for images to extract features and create a dataset. Nearly 10 000 profiles and 30 000 comments from public accounts were manually tagged to create the classification model. The final model had an accuracy rate of 95% on the dataset, allowing for the effective identification of different types of business accounts on Instagram.
- 8 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (2) , 2776-2783
ICON: Instagram Profile Classification Using Image and Natural Language Processing Methods
The use of social media has grown significantly, and businesses are now using these platforms to promote their products and services. To do this, companies have created business accounts on social networks. However, social media platforms can also be a breeding ground for unwanted behaviors such as cyberbullying, sexual content, and promotional comments. To address this issue, a study was conducted to create a system that could classify public accounts on Instagram by analyzing comments, profile pictures, bios, and posts shared by users with business accounts. First, a crawler was developed, and data were collected using this crawler and then anonymized. Next, the collected data were processed using natural language processing (NLP) techniques for text and image processing methods for images to extract features and create a dataset. Nearly 10 000 profiles and 30 000 comments from public accounts were manually tagged to create the classification model. The final model had an accuracy rate of 95% on the dataset, allowing for the effective identification of different types of business accounts on Instagram.
- 1 August 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21540217,p. 33-38
A framework for modeling and implementing QoS-aware load balancing solutions in WiFi hotspots
Access Point (AP) selection in WiFi hotspots is driven by stations and it is based on the measured strongest RSSI (Received Signal Strength Indicator) level: any station connects to the AP that provides the higher physical data rate. Although simple and effective in low crowded scenarios with low-medium traffic load, this strategy performs inefficiently when the number of mobile users is high and their distribution among APs is unbalanced, i.e. when network congestion becomes an issue. Load Balancing (LB) solutions aim at solving this problem by enforcing the connection of stations to the AP having either the smallest number of associated stations or the lowest traffic load. However, LB solutions do not account for traffic priorities or, when they consider them, they do not deal with the joint configuration of QoS (Quality of Service) and LB parameters. In this study we present a framework for modeling, analyzing and designing QoS-aware LB solutions. The proposed framework assumes that stations implement the Enhanced Distributed Channel Access (EDCA) mechanism of the IEEE 802.11e standard. Moreover, in order to make the framework concrete, we assume that the QoS goal is the weighted fair allocation of wireless resources. However, the framework is not restricted to this goal and can be easily extended in order to deal with a different cost function. The proposed framework is validated through simulations in a typical indoor LB scenario. The results show that the model is effective in capturing network performance and in designing LB solutions that account for traffic priorities and the configuration of QoS parameters.
- 1 August 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21540217,p. 33-38
A framework for modeling and implementing QoS-aware load balancing solutions in WiFi hotspots
Access Point (AP) selection in WiFi hotspots is driven by stations and it is based on the measured strongest RSSI (Received Signal Strength Indicator) level: any station connects to the AP that provides the higher physical data rate. Although simple and effective in low crowded scenarios with low-medium traffic load, this strategy performs inefficiently when the number of mobile users is high and their distribution among APs is unbalanced, i.e. when network congestion becomes an issue. Load Balancing (LB) solutions aim at solving this problem by enforcing the connection of stations to the AP having either the smallest number of associated stations or the lowest traffic load. However, LB solutions do not account for traffic priorities or, when they consider them, they do not deal with the joint configuration of QoS (Quality of Service) and LB parameters. In this study we present a framework for modeling, analyzing and designing QoS-aware LB solutions. The proposed framework assumes that stations implement the Enhanced Distributed Channel Access (EDCA) mechanism of the IEEE 802.11e standard. Moreover, in order to make the framework concrete, we assume that the QoS goal is the weighted fair allocation of wireless resources. However, the framework is not restricted to this goal and can be easily extended in order to deal with a different cost function. The proposed framework is validated through simulations in a typical indoor LB scenario. The results show that the model is effective in capturing network performance and in designing LB solutions that account for traffic priorities and the configuration of QoS parameters.