Yazar "Tanyildiz, Hayriye" seçeneğine göre listele

Listeleniyor 1 - 4 / 4
  • Küçük Resim Yok
    Öğe
    A Methodological Study on Fetal Health Classification Using Optimized LightGBM with SMOTE and Optuna-Based Hyperparameter Optimization
    (Turkiye Klinikleri, 2025) Yalcin, Emre; Aslan, Serpil; Tanyildiz, Hayriye
    Objective: Fetal health classification is of great clinical importance as it allows the early detection and management of fetal health problems during pregnancy.This study aims to enhance fetal health classification by integrating Light Gradient Boosting Machine (LightGBM), synthetic minority oversampling technique (SMOTE), and Optuna-based hyperparameter tuning. The goal is to improve classification accuracy, address class imbalance, and optimize model performance in predicting normal, suspicious, and pathological fetal health conditions. Material and Methods: The study utilized the University of California, Irvine Cardiotocography Data Set which contains 2.126 fetal cardiotocography (CTG) records classified into 3 categories. To handle class imbalance, SMOTE is applied. Various machine learning models [Stochastic Gradient Descent Classifier, Gradient Boosting, Decision Tree, Support Vector Machine, K-Nearest Neighbors, and Logistic Regression (LG)] were compared, with LightGBM selected due to its efficiency in structured medical data processing. Optuna is used for hyperparameter tuning. The dataset was split into 80% training and 20% testing, and model performance was assessed using accuracy, precision, recall, and F1-score. Results: The optimized LightGBM+SMOTE+Optuna model achieved 99% accuracy, significantly outperforming the other classifiers. The recall and F1-score for the suspicious and pathological classes improved, reducing the misclassification rates. The confusion matrix confirmed a substantial decrease in errors, demonstrating the model's robustness and reliability. Conclusion: The proposed model successfully enhances fetal health classification accuracy by addressing class imbalance and optimizing hyperparameters. The integration of LightGBM, SMOTE, and Optuna improves model generalization, making it a valuable tool for fetal health assessment. Future studies could explore deep learning approaches to further refine classification performance and support clinical decision-making.
  • Küçük Resim Yok
    Öğe
    Detection of cyber attacks in electric vehicle charging systems using a remaining useful life generative adversarial network
    (Nature Portfolio, 2025) Tanyildiz, Hayriye; Sahin, Canan Batur; Dinler, Ozlem Batur; Migdady, Hazem; Saleem, Kashif; Smerat, Aseel; Abualigah, Laith
    Cybersecurity attacks targeting electric vehicle supply equipment (EVSE) and the broader electric vehicle (EV) ecosystem have become an escalating concern with the increasing adoption of EVs and the growing connectivity of the infrastructure supporting them. The present research aims to contribute to continuing cybersecurity studies on electric vehicle charging stations. In line with this objective, this study proposes the remaining useful life (RUL) approach to demonstrate the potential impact of estimating the remaining time of a cyber attack on EVSE and what revolutionary changes it can bring to cyber security strategies using a generative adversarial network (GAN). By taking a proactive stance, the manuscript will increase security and reduce the economic and reputational losses associated with cyber incidents. Accurate RUL estimates present valuable information about the status of the EVSE infrastructure. Thus, informed decisions on maintenance and crew scheduling are taken. To test the technique's effectiveness, we assess this approach on attack scenarios, including network and host attacks on the EV charger (Electric Vehicle Supply Equipment-EVSE) in idle and charging states. Furthermore, we assess the prediction results of different deep learning models, such as gated recurrent units (GRUs), long short-term memory (LSTM), recurrent neural networks (RNNs), convolution neural networks (CNNs), multi-layer perceptron (MLP), and dense layer integrated with generative adversarial networks (GANs), using mean absolute error (MAE), root mean square error (RMSE), mean squared error (MSE), and R-squared (R2). Afterward, we compare the error measurements with models, such as hybrid GAN-LSTM, GAN-GRU, GAN-RNN, GAN-CNN, GAN-MLP, and GAN-Dense Layer. The GAN-GRU model exhibits the highest accuracy with the lowest MAE (0.0281). On the contrary, the GAN-CNN model displays the best overall performance concerning error consistency and variance explained. According to the results, integrating GAN into these architectures improves predictive accuracy and the model's ability to identify potential attacks in advance and decreases error rates.
  • Küçük Resim Yok
    Öğe
    Explainable tabular deep learning models for antenatal cesarean delivery prediction in multiparous women
    (Bmc, 2026) Yalcin, Emre; Tanyildiz, Hayriye; Aslan, Serpil; Demir, Suleyman Cansun; Sucu, Mete; Uzay, Fatma Islek; Bicer, Ayse
    Background/Objectives Globalincreases in cesarean section (C-section) rates, often exceeding medical necessity, highlight the need for accurate antenatal prediction to support evidence-based birth planning. Reliable prediction of delivery mode is essential for reducing maternal and neonatal morbidity, improving clinical decision-making, and optimizing resource allocation. This study analyzes a publicly available dataset of 460 multiparous women, including 18 obstetric and antenatal variables, published by Yimer and Mekonnen. Methods Deep learning architectures were systematically evaluated for predicting delivery mode in multiparous pregnancies. Classical Multilayer Perceptrons (MLPs) served as baseline models, while modern tabular deep learning methods were assessed as advanced alternatives. Preprocessing included multiple imputation, outlier removal, and class balancing via SMOTE. Feature selection was performed using a hybrid Boruta-clinical expert strategy. Hyperparameters were tuned through Random Search. To improve interpretability, an explainability pipeline integrating SHAP and LIME was incorporated. Results Optimized MLPs produced modest performance gains, but dedicated tabular models demonstrated clear superiority. TabNet achieved the highest performance, with an ROC-AUC of 0.79 and a PR-AUC of 0.74, attributed to its attention and masking mechanisms and robust handling of minority classes. TabPFN and CBAM-MLP yielded stable and balanced results, whereas FT-Transformer showed competitive yet comparatively moderate accuracy. Conclusions The findings demonstrate that modern tabular deep learning approaches, particularly TabNet, surpass baseline MLP architectures in terms of accuracy, explainability, and clinical applicability for predicting C-section in multiparous women. This study presents the first comprehensive and explainable comparison of tabular deep learning models tailored to multiparous pregnancies, combining hybrid Boruta-expert feature selection with SHAP and LIME interpretability. TabNet emerges as the most promising candidate for integration into clinical decision support systems, contributing substantially to Al-driven strategies for addressing rising global C-section rates.
  • Küçük Resim Yok
    Öğe
    Improving Deceptive Patch Solutions Using Novel Deep Learning-Based Time Analysis Model for Industrial Control Systems
    (Mdpi, 2024) Tanyildiz, Hayriye; Sahin, Canan Batur; Dinler, Ozlem Batur
    Industrial control systems (ICSs) are critical components automating the processes and operations of electromechanical systems. These systems are vulnerable to cyberattacks and can be the targets of malicious activities. With increased internet connectivity and integration with the Internet of Things (IoT), ICSs become more vulnerable to cyberattacks, which can have serious consequences, such as service interruption, financial losses, and security hazards. Threat actors target these systems with sophisticated attacks that can cause devastating damage. Cybersecurity vulnerabilities in ICSs have recently led to increasing cyberattacks and malware exploits. Hence, this paper proposes to develop a security solution with dynamic and adaptive deceptive patching strategies based on studies on the use of deceptive patches against attackers in industrial control systems. Within the present study's scope, brief information on the adversarial training method and window size manipulation will be presented. It will emphasize how these methods can be integrated into industrial control systems and how they can increase cybersecurity by combining them with deceptive patch solutions. The discussed techniques represent an approach to improving the network and system security by making it more challenging for attackers to predict their targets and attack methods. The acquired results demonstrate that the suggested hybrid method improves the application of deception to software patching prediction, reflecting enhanced patch security.