WoS İndeksli Yayınlar Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12416/8653
Browse
4 results
Search Results
Conference Object Citation - WoS: 1Citation - Scopus: 1Dengesiz Epilepsi Veri Seti İçin Sınıflandırmada Farklı SMOTE Yöntemlerinin Etkileri(Institute of Electrical and Electronics Engineers Inc., 2025) Calis, Ahmet Gokay; Ergezer, HalitIn this study, the effects of different SMOTE methods on machine learning algorithms for the imbalanced epilepsy dataset were investigated. After filtering, the imbalanced dataset was balanced with 5 different SMOTE methods and classified with various machine learning algorithms. Coarse-K-Nearest Neighbor, Bagged Trees, and Artificial Neural Networks models were evaluated in epilepsy detection. The performance of these different models was compared with Matthews Correlation Coefficient (MCC) and F1 Score metrics. The results showed that the Borderline-SMOTE algorithm had the highest F1 Score and MCC values among all machine learning algorithms. © 2025 Elsevier B.V., All rights reserved.Conference Object Citation - WoS: 1Evaluation of Features Used in Electromyography Classification(Ieee, 2021) Ergezer, Halit; Alguner, Ayber ErayClassification of electromyography (EMG) signals using machine learning has been studied for a long time. Today, this classification is tried to be made more accurate, fast and applicable by using the methods developed. However, beside this effort, it is suspected that researchers are using features without taking into account the effects on the classification performance, but often by influence of other researches. From this point of view, the effects of some features used in studies published in recent years on classification performance were tested and the results obtained were shared. In the experiments performed using a common method support vector machine (SVM), it was found that increasing the number of features does not always provide an increase in performance, even in some cases, it causes a decrease in accuracy rates.Conference Object Design and Implementation of Visual Simultaneous Localization and Mapping (Vslam) Navigation System(Ieee, 2021) Ergezer, Halit; Bekcan, ArdaIt is very important to guess the location of the redetected objects and loop closures with the visual simultaneous localization and mapping system (VSLAM), one of the biggest problems of a mobile robot. VSLAM makes it possible to eliminate and/or reduce these applications' errors and realize or improve the robot's direction and position correctly by creating a map of the environment. This study aims to achieve an autonomous indoor/outdoor navigation of a ground robot using VSLAM algorithm in an unknown environment using a monocular camera. In this context, the theoretical information was tested in real-world conditions. Performance of localization and loop closing were compared based on the results obtained by experimentsArticle Citation - WoS: 1Citation - Scopus: 2Classification of Low Probability of Intercept Radar Waveforms Using Gabor Wavelets(Gazi Univ, Fac Engineering Architecture, 2021) Ergezer, HalitLow Probability of Intercept (LPI Radar) is a class of radar with specific technical characteristics that make it very difficult to intercept with electronic support systems and radar warning receivers. Because of their properties as low power, variable frequency, wide bandwidth, LPI radar waveforms are difficult to intercept by ESM systems. In recent years, studies on the classification of waveforms used by these types of radar have been accelerated. In this study, Time-Frequency Images (TFI) has been obtained from the LPI radars waveforms by using Choi-Williams Distribution method. From these images, feature vectors have been generated using Gabor Wavelet transform. In contrast to many methods in the literature, waveform classification has been performed by directly comparing the feature vectors obtained without using any machine learning method. With the method we propose, classification accuracies were obtained at intervals of 2 dB between -20 dB and 10 dB and performed at reasonable classification accuracy rates up to -8 dB SNR value. Better results than the best reported in the literature were obtained for some signal types. The results obtained for all waveform types are given in comparison with the results of the existing methods in the literature.