Mühendislik Fakültesi / Faculty of Engineering
Permanent URI for this collectionhttps://hdl.handle.net/11727/1401
Browse
7 results
Search Results
Item Obstructive Sleep Apnea Classification with Artificial Neural Network Based On Two Synchronic Hrv Series(2015) Aksahin, Mehmet; Erdamar, Aykut; Firat, Hikmet; Ardic, Sadik; Erogul, Osman; 0000-0001-8588-480X; AAA-6844-2019In the present study, "obstructive sleep apnea (OSA) patients" and "non-OSA patients" were classified into two groups using with two synchronic heart rate variability (HRV) series obtained from electrocardiography (ECG) and photoplethysmography (PPG) signals. A linear synchronization method called cross power spectrum density (CPSD), commonly used on HRV series, was performed to obtain high-quality signal features to discriminate OSA from controls. To classify simultaneous sleep ECG and PPG signals recorded from OSA and non-OSA patients, various feed forward neural network (FFNN) architectures are used and mean relative absolute error (MRAE) is applied on FFNN results to show affectivities of developed algorithm. The FFNN architectures were trained with various numbers of neurons and hidden layers. The results show that HRV synchronization is directly related to sleep respiratory signals. The CPSD of the HRV series can confirm the clinical diagnosis; both groups determined by an expert physician can be 99% truly classified as a single hidden-layer FFNN structure with 0.0623 MRAE, in which the maximum and phase values of the CPSD curve are assigned as two features. In future work, features taken from different physiological signals can be added to define a single feature that can classify apnea without error.Item Using Machine Learning Methods in Early Diagnosis of Breast Cancer(2021) Erkal, Begum; Ayyildiz, Tulin Ercelebi; https://orcid.org/0000-0002-7372-0223; JBI-6492-2023Breast cancer is one of the most important health diseases to be treated in the world, and it is a subject that has a wide place in research subjects. In this study, results are provided by using seven different machine learning techniques for the classification of breast cancer. In order to obtain better results, the preprocessing method was applied. As a result, when compared with some studies in the literature, it was observed that the general performance of some of the methods improved. In the experimental results, BayesNet was found to be the best classification method with an accuracy rate of 97.13%.Item Applications of Deep Learning Techniques to Wood Anomaly Detection(2022) Celik, Yaren; Guney, Selda; Dengiz, Berna; Xu, J; Altiparmak, F.; Hassan, MHA; Marquez, FPGWood products and structures have an important place in today's industry. They are widely used in many fields. However, there are various difficulties in production systems where wood raw material is under many processes. Some difficulty and complexity of production processes result in high variability of raw materials such as a wide range of visible structural defects that must be checked by specialists on line or of line. These issues are not only difficult and biased in manual processes, but also less effective and misleading. To overcome the drawbacks of the manual quality control processes, machine vision-based inspection systems are in great of interest recently for quality control applications. In this study, the wood anomaly has been detected by using deep learning. As it will be a distinction-based method on image processing, the Convolution Neural Network (CNN), which is one of the most suitable methods, has been used for anomaly detection. In addition, it will be tried to obtain the most suitable one among different CNN architectures such as ShuffleNet, AlexNet, GoogleNet for the problem. MobileNet, SqueezeNet, GoogleNet, ShuffleNet among considered methods show promising results in classifying normal and abnormal wood products.Item Detection of multiple sclerosis from photic stimulation EEG signals(2021) Karaca, Busra Kubra; Aksahin, Mehmet Feyzi; Ocal, RuhsenBackground: Multiple Sclerosis (MS) is characterized as a chronic, autoimmune and inflammatory disease of the central nervous system. Early diagnosis of MS is of great importance for the treatment and course of the disease. In addition to the many methods, cost-effective and non-invasive electroencephalogram signals may contribute to the pre-diagnosis of MS. Objectives: The aim of this paper is to classify male subjects who have MS and who are healthy control using photic stimulation electroencephalogram signals. Methods: Firstly the continuous wavelet transformation (CWT) method was applied to electroencephalogram signals under photic stimulation with 5Hz, 10Hz, 15Hz, 20Hz, and 25Hz frequencies. The sum, maximum, minimum and standard deviation values of absolute CWT coefficients, corresponding to "1-4 Hz" and "4-13 Hz" frequency ranges, were extracted in each stimulation frequency region. The ratios of these values obtained from the frequency ranges "1-4Hz" and "4-13Hz" was decided as features. Finally, various machine learning classifiers were evaluated to test the effectivity of determined features. Results: Consequently, the overall accuracy, sensitivity, specificity and positive predictive value of the proposed algorithm were 80 %, 72.7 %, 88.9 %, and 88.9 %, respectively by using the Ensemble Subspace k-NN classifier algorithm. Conclusions: The results showed how photic stimulation electroencephalogram signals can contribute to the prediagnosis of MS.Item Obesity Level Estimation based on Machine Learning Methods and Artificial Neural Networks(2021) Celik, Yaren; Guney, Selda; Dengiz, BernaObesity is a growing societal and public health problem starting from 1980 that needs more attention. For this reason, new studies are emerging day by day, including those looking for obesity in children, especially the impact factors, and how to predict the emergence of the situation under these factors. In this study, different classification methods were applied for the estimation of obesity levels. Based on the evaluation criteria, the results were compared for different machine learning methods. When the Cubic SVM method was applied by selecting the appropriate features specific to the problem, 97.8% accuracy was obtained.Item Wi-Fi Based Indoor Positioning System with Using Deep Neural Network(2020) Guney, Selda; Erdogan, Alperen; Aktas, Melih; Ergun, MertIndoor positioning is one of the major challenges for the future large-scale technologies. Nowadays, it has become an attractive research subject due to growing demands on it. Several algorithms and techniques have been developed over the decades. One of the most cost-effective technique is Wi-Fi-based positioning systems. This technique is infrastructure-free and able to use existing wireless access points in public or private areas. These systems aim to classify user's location according to pre-defined set of grids. However, Wi-Fi signals could be affected by interference, blockage of walls and multipath effect which increases error of classification. In this study Deep Neural Networks and conventional machine learning classifiers are utilized to classify 22 squared grids which represent locations. Five primary Wireless Access Points (WAPs) were mounted indoor environment and 177 secondary WAPs are observed by Wi-Fi module. Dataset was created with using five primary and 177 secondary WAPs. The performance of proposed method was tested using Deep Neural Networks and machine learning classifiers. The results show that Deep Neural Network present the best performance as compared to machine learning classifiers. 95.45% accuracy was achieved by using five primary WAPs and 97.27% accuracy was achieved by using five primary and 177 secondary WAPs together for Deep Neural Network.Item Sparsity-driven weighted ensemble classifier(2018) Erdem, Hamit; Ozgur, Atilla; Nar, FatihIn this study, a novel sparsity-driven weighted ensemble classifier (SDWEC) that improves classification accuracy and minimizes the number of classifiers is proposed. Using pre-trained classifiers, an ensemble in which base classifiers votes according to assigned weights is formed. These assigned weights directly affect classifier accuracy. In the proposed method, ensemble weights finding problem is modeled as a cost function with the following terms: (a) a data fidelity term aiming to decrease misclassification rate, (b) a sparsity term aiming to decrease the number of classifiers, and (c) a non-negativity constraint on the weights of the classifiers. As the proposed cost function is non-convex thus hard to solve, convex relaxation techniques and novel approximations are employed to obtain a numerically efficient solution. Sparsity term of cost function allows trade-off between accuracy and testing time when needed. The efficiency of SDWEC was tested on 11 datasets and compared with the state-of-the art classifier ensemble methods. The results show that SDWEC provides better or similar accuracy levels using fewer classifiers and reduces testing time for ensemble.