Süha Tuna
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Süha Tuna has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Metin Demiralp
Istanbul Technical University
3 shared publications - Burcu Tunga
Department of Mathematics Engineering, Faculty of Science and Letters, İstanbul Technical University, 34469 Maslak, İstanbul, Türkiye
2 shared publications Ayse Cirakoglu
Istanbul University-Cerrahpasa
1 shared publication- Ayse Cirakoglu
Department of Medical Biology, Cerrahpasa Faculty of Medicine, Istanbul University-Cerrahpasa, Istanbul, Trkiye
1 shared publication Behcet Ugur Toreyin
Applied Informatics Department, Informatics Institute, Istanbul Technical University, Istanbul, Turkey
1 shared publication- Bugra Eyidogan1 shared publication
Latest Publications
- Open Access
- 25 June 2026
- Published by Elsevier in SSRN Electronic Journal
MechCube: Tensor Decomposition Based Surrogate Modeling of Mode-I Stress Intensity Factors
Repeated evaluation of crack-tip stress intensity factors in finite-element- based fracture analysis becomes computationally demanding in large para- metric engineering studies. This work proposes MechCube, a tensor- decomposition-based surrogate modeling framework for estimating the mode-separated Mode-I stress intensity factor component KI from finite element crack-tip stress–strain fields. Local stresses and strains extracted from simulations of edge-cracked plates, internally cracked plates, and cracked plates containing holes are assembled into fixed-size tensors through a bijective crack-tip tensorization procedure. The resulting representa- tions are processed using Enhanced Multivariance Products Representation (EMPR), High Dimensional Model Representation (HDMR), CANDECOM- P/PARAFAC (CP), and Tucker decomposition to construct compact in- teraction maps for convolutional neural network regression. The proposed workflow provides a structured representation-learning strategy for high- dimensional crack-tip fields, while the target application is repeated KI evaluation in finite-element-based fracture analysis under varying crack lengths, crack angles, loading levels, and geometric configurations. The sur- rogate target is the mode-separated Mode-I component KI, taken directly from the finite element interaction-integral output. Experiments on more than 1.2 × 105 numerically stable finite element samples spanning six crack configurations show that Tucker-based representations provide the most ac- curate and robust surrogate estimates, outperforming raw-field inputs and alternative tensor decomposition strategies while also improving convergence stability. In addition, the decomposed representations reveal interpretable crack-tip interaction patterns that remain mechanically meaningful across the benchmark cases. These results indicate that tensor-decomposition- based surrogate modeling can support scalable and interpretable crack-tip parameter evaluation in data-intensive engineering workflows.
- 25 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMechCube: Tensor Decomposition Based Surrogate Modeling of Mode-I Stress Intensity Factors
Repeated evaluation of crack-tip stress intensity factors in finite-element- based fracture analysis becomes computationally demanding in large para- metric engineering studies. This work proposes MechCube, a tensor- decomposition-based surrogate modeling framework for estimating the mode-separated Mode-I stress intensity factor component KI from finite element crack-tip stress–strain fields. Local stresses and strains extracted from simulations of edge-cracked plates, internally cracked plates, and cracked plates containing holes are assembled into fixed-size tensors through a bijective crack-tip tensorization procedure. The resulting representa- tions are processed using Enhanced Multivariance Products Representation (EMPR), High Dimensional Model Representation (HDMR), CANDECOM- P/PARAFAC (CP), and Tucker decomposition to construct compact in- teraction maps for convolutional neural network regression. The proposed workflow provides a structured representation-learning strategy for high- dimensional crack-tip fields, while the target application is repeated KI evaluation in finite-element-based fracture analysis under varying crack lengths, crack angles, loading levels, and geometric configurations. The sur- rogate target is the mode-separated Mode-I component KI, taken directly from the finite element interaction-integral output. Experiments on more than 1.2 × 105 numerically stable finite element samples spanning six crack configurations show that Tucker-based representations provide the most ac- curate and robust surrogate estimates, outperforming raw-field inputs and alternative tensor decomposition strategies while also improving convergence stability. In addition, the decomposed representations reveal interpretable crack-tip interaction patterns that remain mechanically meaningful across the benchmark cases. These results indicate that tensor-decomposition- based surrogate modeling can support scalable and interpretable crack-tip parameter evaluation in data-intensive engineering workflows.
- 30 November 2025
- Published by Kocaeli Journal of Science and Engineering in Kocaeli Journal of Science and Engineering
- Vol. 8 (2) , 121-132
Refining Sparse Coding Dictionaries Using High Dimensional Model Representation for Hyperspectral Imagery
Hyperspectral imaging, known for its rich spectral and spatial details, finds applications in remote sensing, biomedical engineering, and quality control. Nonetheless, the high dimensionality and large data volume of hyperspectral images create substantial challenges in efficient processing and classification. Sparse coding-based techniques are widely employed to tackle these challenges. Therefore, an appropriate dictionary should be constructed to improve the efficacy of the sparse coding-based methods. This study introduces a dictionary refinement method that enhances sparse coding-based classification by exploiting an efficient denoising and decorrelation technique named High Dimensional Model Representation. This technique decomposes the 3-D hyperspectral data into manageable components, effectively reducing noise and correlations. Then, a refined dictionary is acquired by using random spectral signals of the hyperspectral data under consideration. The sparse coding-based classifier adopting the refined dictionary is exploited to improve the classification accuracy. Experimental results on widely used HS datasets show that the proposed method significantly boosts classification accuracy. This method leverages the benefits of denoising and decorrelation of the High Dimensional Model Representation method to generate a refined dictionary and provide a robust and efficient solution for hyperspectral image classification.
- 30 November 2025
- Published by Kocaeli Journal of Science and Engineering in Kocaeli Journal of Science and Engineering
- Vol. 8 (2) , 121-132
Refining Sparse Coding Dictionaries Using High Dimensional Model Representation for Hyperspectral Imagery
Hyperspectral imaging, known for its rich spectral and spatial details, finds applications in remote sensing, biomedical engineering, and quality control. Nonetheless, the high dimensionality and large data volume of hyperspectral images create substantial challenges in efficient processing and classification. Sparse coding-based techniques are widely employed to tackle these challenges. Therefore, an appropriate dictionary should be constructed to improve the efficacy of the sparse coding-based methods. This study introduces a dictionary refinement method that enhances sparse coding-based classification by exploiting an efficient denoising and decorrelation technique named High Dimensional Model Representation. This technique decomposes the 3-D hyperspectral data into manageable components, effectively reducing noise and correlations. Then, a refined dictionary is acquired by using random spectral signals of the hyperspectral data under consideration. The sparse coding-based classifier adopting the refined dictionary is exploited to improve the classification accuracy. Experimental results on widely used HS datasets show that the proposed method significantly boosts classification accuracy. This method leverages the benefits of denoising and decorrelation of the High Dimensional Model Representation method to generate a refined dictionary and provide a robust and efficient solution for hyperspectral image classification.
- Open Access
- 30 June 2024
- Published by Plusbase Akademi Organizasyon ve Danismanlik Ltd.Sti. in International Journal of Applied Methods in Electronics and Computers
- Vol. 12 (2) , 40-47
Enhancing Travel Experience: Predicting Flight Delays for Informed Journey Planning
Flight delays pose significant inconveniences for travelers, potentially causing missed connections, schedule adjustments, and time wastage. This study presents a machine-learning driven approach to mitigate these challenges by developing an application that predicts flight delays, empowering passengers with insights to minimize travel disruptions. Leveraging diverse machine learning algorithms and datasets from the United States Department of Transportation and the National Oceanic and Atmospheric Administration Service, our model aids travelers in making informed decisions by suggesting optimal flight times and carriers based on historical flight data and weather conditions. Addressing the issue of imbalanced data, we explore techniques such as Synthetic Minority Over-sampling Technique (SMOTE) and under-sampling. Our comparative analysis highlights the superior performance of Light Gradient Boosting Machine (L-GBM) in predicting flight delays. With an F1-score of and an AUC value of , our study offers a promising solution to enhance passenger experiences through improved flight recommendations.
- 30 June 2024
- Published by Plusbase Akademi Organizasyon ve Danismanlik Ltd.Sti. in International Journal of Applied Methods in Electronics and Computers
- Vol. 12 (2) , 40-47
Open AccessEnhancing Travel Experience: Predicting Flight Delays for Informed Journey Planning
Flight delays pose significant inconveniences for travelers, potentially causing missed connections, schedule adjustments, and time wastage. This study presents a machine-learning driven approach to mitigate these challenges by developing an application that predicts flight delays, empowering passengers with insights to minimize travel disruptions. Leveraging diverse machine learning algorithms and datasets from the United States Department of Transportation and the National Oceanic and Atmospheric Administration Service, our model aids travelers in making informed decisions by suggesting optimal flight times and carriers based on historical flight data and weather conditions. Addressing the issue of imbalanced data, we explore techniques such as Synthetic Minority Over-sampling Technique (SMOTE) and under-sampling. Our comparative analysis highlights the superior performance of Light Gradient Boosting Machine (L-GBM) in predicting flight delays. With an F1-score of and an AUC value of , our study offers a promising solution to enhance passenger experiences through improved flight recommendations.
- 4 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Computational Biology and Bioinformatics
- Vol. 20 (5) , 2959-2969
Gene Teams are on the Field: Evaluation of Variants in Gene-Networks Using High Dimensional Modelling
In medical genetics, each genetic variant is evaluated as an independent entity regarding its clinical importance. However, in most complex diseases, variant combinations in specific gene networks, rather than the presence of a particular single variant, predominates. In the case of complex diseases, disease status can be evaluated by considering the success level of a team of specific variants. We propose a high dimensional modelling based method to analyse all the variants in a gene network together, which we name “Computational Gene Network Analysis” (CoGNA).To evaluate our method, we selected two gene networks, mTOR and TGF- $\beta$ . For each pathway, we generated 400 control and 400 patient group samples. mTOR and TGF- $\beta$ pathways contain 31 and 93 genes of varying sizes, respectively. We produced Chaos Game Representation images for each gene sequence to obtain 2-D binary patterns. These patterns were arranged in succession, and a 3-D tensor structure was achieved for each gene network. Features for each data sample were acquired by exploiting Enhanced Multivariance Products Representation to 3-D data. Features were split as training and testing vectors. Training vectors were employed to train a Support Vector Machines classification model. We achieved more than $96\%$ and $99\%$ classification accuracies for mTOR and TGF- $\beta$ networks, respectively, using a limited amount of training samples.
- 4 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Computational Biology and Bioinformatics
- Vol. 20 (5) , 2959-2969
Gene Teams are on the Field: Evaluation of Variants in Gene-Networks Using High Dimensional Modelling
In medical genetics, each genetic variant is evaluated as an independent entity regarding its clinical importance. However, in most complex diseases, variant combinations in specific gene networks, rather than the presence of a particular single variant, predominates. In the case of complex diseases, disease status can be evaluated by considering the success level of a team of specific variants. We propose a high dimensional modelling based method to analyse all the variants in a gene network together, which we name “Computational Gene Network Analysis” (CoGNA).To evaluate our method, we selected two gene networks, mTOR and TGF- $\beta$ . For each pathway, we generated 400 control and 400 patient group samples. mTOR and TGF- $\beta$ pathways contain 31 and 93 genes of varying sizes, respectively. We produced Chaos Game Representation images for each gene sequence to obtain 2-D binary patterns. These patterns were arranged in succession, and a 3-D tensor structure was achieved for each gene network. Features for each data sample were acquired by exploiting Enhanced Multivariance Products Representation to 3-D data. Features were split as training and testing vectors. Training vectors were employed to train a Support Vector Machines classification model. We achieved more than $96\%$ and $99\%$ classification accuracies for mTOR and TGF- $\beta$ networks, respectively, using a limited amount of training samples.
- Open Access
- 16 February 2023
- Published by Springer Nature in Research Square
Gene Teams are on the Field: Evaluation of Variants in Gene-Networks Using High Dimensional Modelling
Variation is a key concept in every biological aspect, particularly in medical genetics. In this field, each genetic variant is evaluated mostly as an independent entity in respect of its clinical importance. This approach may be sufficient to detect the pathogenic variants in single-gene disorders. However, in most of the complex diseases, the combination of the variants in specific gene networks, rather than the presence of a certain single variant, predominates. Therefore, in case of a complex disease, the disease status can be evaluated by considering it as the success level of a team composed of certain variants. To assess the feasibility of this approach, we tested the effectiveness of high-dimensional modelling of gene network-restricted variants in distinguishing a disease status. To evaluate the proposed method, we selected two gene networks, mTOR and TGF-β. For each pathway, we generated 400 control and 400 patient group samples. The considered mTOR and TGF-β pathways contain 31 and 93 genes of varying sizes, respectively. We generated Chaos Game Representation images for each gene sequence to obtain 2-D binary patterns. Produced patterns were arranged in succession, and a 3-D tensor structure was achieved for each gene network. Features for each data sample were acquired by exploiting Enhanced Multivariance Products Representation to 3-D data. The features were split as training and testing vectors. The training vectors were employed to train a Support Vector Machines classification model. We managed to achieve more than 96% and 99% classification accuracies for mTOR and TGF-β networks, respectively, using a limited amount of training samples.
- 16 February 2023
- Published by Springer Nature in Research Square
Open AccessGene Teams are on the Field: Evaluation of Variants in Gene-Networks Using High Dimensional Modelling
Variation is a key concept in every biological aspect, particularly in medical genetics. In this field, each genetic variant is evaluated mostly as an independent entity in respect of its clinical importance. This approach may be sufficient to detect the pathogenic variants in single-gene disorders. However, in most of the complex diseases, the combination of the variants in specific gene networks, rather than the presence of a certain single variant, predominates. Therefore, in case of a complex disease, the disease status can be evaluated by considering it as the success level of a team composed of certain variants. To assess the feasibility of this approach, we tested the effectiveness of high-dimensional modelling of gene network-restricted variants in distinguishing a disease status. To evaluate the proposed method, we selected two gene networks, mTOR and TGF-β. For each pathway, we generated 400 control and 400 patient group samples. The considered mTOR and TGF-β pathways contain 31 and 93 genes of varying sizes, respectively. We generated Chaos Game Representation images for each gene sequence to obtain 2-D binary patterns. Produced patterns were arranged in succession, and a 3-D tensor structure was achieved for each gene network. Features for each data sample were acquired by exploiting Enhanced Multivariance Products Representation to 3-D data. The features were split as training and testing vectors. The training vectors were employed to train a Support Vector Machines classification model. We managed to achieve more than 96% and 99% classification accuracies for mTOR and TGF-β networks, respectively, using a limited amount of training samples.
- 13 December 2022
- Published by Taylor & Francis in International Journal of Remote Sensing
- Vol. 43 (19-24) , 6899-6920
An efficient feature extraction approach for hyperspectral images using Wavelet High Dimensional Model Representation
Hyperspectral (HS) Imagery helps to capture information using specialized sensors to extract detailed data at numerous narrow wavelengths. Hyperspectral imaging provides both spatial and spectral characteristics of regions or objects for subsequent analysis. Unfortunately, various noise sources decrease the interpretability of these images as well as the correlation between neighbouring pixels, hence both reduce the classification performance. This study focuses on developing an ensemble algorithm that enables to denoise the spectral signals while decorrelating the spatio-spectral features concurrently. The developed method is called Wavelet High Dimensional Model (W-HDMR) and combines High Dimensional Model Representation (HDMR) with the Discrete Wavelet Transform (DWT). Through W-HDMR, denoised and decorrelated features are extracted from the HS cubes. HDMR decorrelates each dimension in HS data while DWT denoises the spectral signals. The classification performance of W-HDMR as a new feature extraction technique for HS images is assessed by exploiting a Support Vector Machines algorithm. The results indicate that the proposed W-HDMR method is an efficient feature extraction technique and is considered an adequate tool in the HS classification problem.
- 13 December 2022
- Published by Taylor & Francis in International Journal of Remote Sensing
- Vol. 43 (19-24) , 6899-6920
An efficient feature extraction approach for hyperspectral images using Wavelet High Dimensional Model Representation
Hyperspectral (HS) Imagery helps to capture information using specialized sensors to extract detailed data at numerous narrow wavelengths. Hyperspectral imaging provides both spatial and spectral characteristics of regions or objects for subsequent analysis. Unfortunately, various noise sources decrease the interpretability of these images as well as the correlation between neighbouring pixels, hence both reduce the classification performance. This study focuses on developing an ensemble algorithm that enables to denoise the spectral signals while decorrelating the spatio-spectral features concurrently. The developed method is called Wavelet High Dimensional Model (W-HDMR) and combines High Dimensional Model Representation (HDMR) with the Discrete Wavelet Transform (DWT). Through W-HDMR, denoised and decorrelated features are extracted from the HS cubes. HDMR decorrelates each dimension in HS data while DWT denoises the spectral signals. The classification performance of W-HDMR as a new feature extraction technique for HS images is assessed by exploiting a Support Vector Machines algorithm. The results indicate that the proposed W-HDMR method is an efficient feature extraction technique and is considered an adequate tool in the HS classification problem.
- Open Access
- 23 November 2022
- Published by Elsevier in Engineering Science and Technology, an International Journal
- Vol. 37
Low-cost VIS/NIR range hand-held and portable photospectrometer and evaluation of machine learning algorithms for classification performance
- 23 November 2022
- Published by Elsevier in Engineering Science and Technology, an International Journal
- Vol. 37
Open AccessLow-cost VIS/NIR range hand-held and portable photospectrometer and evaluation of machine learning algorithms for classification performance
- 12 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 59 (11) , 9569-9584
Iterative Enhanced Multivariance Products Representation for Effective Compression of Hyperspectral Images
Effective compression of hyperspectral (HS) images is essential due to their large data volume. Since these images are high dimensional, processing them is also another challenging issue. In this work, an efficient lossy HS image compression method based on enhanced multivariance products representation (EMPR) is proposed. As an efficient data decomposition method, EMPR enables us to represent the given multidimensional data with lower-dimensional entities. EMPR, as a finite expansion with relevant approximations, can be acquired by truncating this expansion at certain levels. Thus, EMPR can be utilized as a highly effective lossy compression algorithm for hyper spectral images. In addition to these, an efficient variety of EMPR is also introduced in this article, in order to increase the compression efficiency. The results are benchmarked with several state-of-the-art lossy compression methods. It is observed that both higher peak signal-to-noise ratio values and improved classification accuracy are achieved from EMPR-based methods.
- 12 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 59 (11) , 9569-9584
Iterative Enhanced Multivariance Products Representation for Effective Compression of Hyperspectral Images
Effective compression of hyperspectral (HS) images is essential due to their large data volume. Since these images are high dimensional, processing them is also another challenging issue. In this work, an efficient lossy HS image compression method based on enhanced multivariance products representation (EMPR) is proposed. As an efficient data decomposition method, EMPR enables us to represent the given multidimensional data with lower-dimensional entities. EMPR, as a finite expansion with relevant approximations, can be acquired by truncating this expansion at certain levels. Thus, EMPR can be utilized as a highly effective lossy compression algorithm for hyper spectral images. In addition to these, an efficient variety of EMPR is also introduced in this article, in order to increase the compression efficiency. The results are benchmarked with several state-of-the-art lossy compression methods. It is observed that both higher peak signal-to-noise ratio values and improved classification accuracy are achieved from EMPR-based methods.
- 21 February 2017
- Published by Springer Nature in Journal of Mathematical Chemistry
- Vol. 55 (6) , 1278-1300
Zero interval limit perturbation expansion for the spectral entities of Hilbert-Schmidt operators combined with most dominant spectral component extraction: convergence and confirmative implementations
- 21 February 2017
- Published by Springer Nature in Journal of Mathematical Chemistry
- Vol. 55 (6) , 1278-1300
Zero interval limit perturbation expansion for the spectral entities of Hilbert-Schmidt operators combined with most dominant spectral component extraction: convergence and confirmative implementations
- 17 February 2017
- Published by Springer Nature in Journal of Mathematical Chemistry
- Vol. 55 (6) , 1253-1277
Zero interval limit perturbation expansion for the spectral entities of Hilbert–Schmidt operators combined with most dominant spectral component extraction: formulation and certain technicalities
- 17 February 2017
- Published by Springer Nature in Journal of Mathematical Chemistry
- Vol. 55 (6) , 1253-1277
Zero interval limit perturbation expansion for the spectral entities of Hilbert–Schmidt operators combined with most dominant spectral component extraction: formulation and certain technicalities