Musa Peker

A new approach for automatic sleep scoring

Automatic classification of sleep stages is one of the most important methods used for diagnostic procedures in psychiatry and neurology. This method, which has been developed by sleep specialists, is a time-consuming and difficult process. Generally, electroencephalogram (EEG) signals are used in sleep scoring. In this study, a new complex classifier-based approach is presented for automatic sleep scoring using EEG signals. In this context, complex-valued methods were utilized in the feature selection and classification stages. In the feature selection stage, features of EEG data were extracted with the help of a dual tree complex wavelet transform (DTCWT). In the next phase, five statistical features were obtained. These features are classified using complex-valued neural network (CVANN) algorithm. The Taguchi method was used in order to determine the effective parameter values in this CVANN. The aim was to develop a stable model involving parameter optimization. Different statistical parameters were utilized in the evaluation phase. Also, results were obtained in terms of two different sleep standards. In the study in which a 2nd level DTCWT and CVANN hybrid model was used, 93.84% accuracy rate was obtained according to the Rechtschaffen & Kales (R&K) standard, while a 95.42% accuracy rate was obtained according to the American Academy of Sleep Medicine (AASM) standard. Complex-valued classifiers were found to be promising in terms of the automatic sleep scoring and EEG data.

A decision support system to improve medical diagnosis using a combination of k-medoids clustering based attribute weighting and SVM

The use of machine learning tools has become widespread in medical diagnosis. The main reason for this is the effective results obtained from classification and diagnosis systems developed to help medical professionals in the diagnosis phase of diseases. The primary objective of this study is to improve the accuracy of classification in medical diagnosis problems. To this end, studies were carried out on 3 different datasets. These datasets are heart disease, Parkinson’s disease (PD) and BUPA liver disorders. Key feature of these datasets is that they have a linearly non-separable distribution. A new method entitled k-medoids clustering-based attribute weighting (kmAW) has been proposed as a data preprocessing method. The support vector machine (SVM) was preferred in the classification phase. In the performance evaluation stage, classification accuracy, specificity, sensitivity analysis, f-measure, kappa statistics value and ROC analysis were used. Experimental results showed that the developed hybrid system entitled kmAWu2009+u2009SVM gave better results compared to other methods described in the literature. Consequently, this hybrid intelligent system can be used as a useful medical decision support tool.

An efficient sleep scoring system based on EEG signal using complex-valued machine learning algorithms

Sleep staging is a significant step in the diagnosis and treatment of sleep disorders. Sleep scoring is a time-consuming and difficult process. Given that sleep scoring requires expert knowledge, it is generally undertaken by sleep experts. In this study, a new hybrid machine learning method consisting of complex-valued nonlinear features (CVNF) and a complex-valued neural network (CVANN) has been presented for automatic sleep scoring using single channel electroencephalography (EEG) signals. First of all, we should note that in this context, nine nonlinear features have been obtained as those are often preferred for the classification of EEG signals. These obtained features were then converted into a complex-valued number format using a phase encoding method. In this way, a new complex-valued feature set was obtained for sleep scoring. The obtained attributes have been presented as input to the CVANN algorithm. We have used a number of different statistical parameters during the evaluation process. The results that have been obtained are based on two sleep standards: Rechtschaffen & Kales (RK a 93.84% accuracy rate was obtained according to the AASM standard using the proposed method. We therefore observed that the proposed method is promising in terms of the sleep scoring. In this study, a new hybrid machine learning method is presented for automatic sleep scoring using EEG signals.In this study, a new complex-valued feature set was obtained for sleep scoring.Complex-valued methods gave good results in classification of EEG data and automatic sleep scoring.With a single channel EEG, this method can reach quite a similar performance with the sleep expert.As part of the study, the behaviour of non-linear features in different sleep stages has been examined.

Signal detection based on empirical mode decomposition and Teager---Kaiser energy operator and its application to P and S wave arrival time detection in seismic signal analysis

Determining P and S wave arrival times while minimizing noise is a major problem in seismic signal analysis. Precise determination of earthquake onset arrival timing, determination of earthquake magnitude, and calculation of other parameters that can be used to make more accurate seismic maps are possible with the detection of these waves. Experts try to determine these waves by manual analysis. But this process is time-consuming and painful. In this study, a new method that enables the determination of P and S wave arrival times in noisy recordings is recommended. This method is based on the hybrid usage of empirical mode decomposition and Teager–Kaiser energy operator algorithms. The results show that the proposed system gives effective results in the automatic detection of P and S wave arrival times. Promisingly, the recommended system might serve as a novel and powerful candidate for the effective detection of P and S wave arrival time.

Karma Taguchi ve yüzey yanıt yöntemi kullanılarak astm b574 (hastelloy c-22)'in tornalanması esnasındaki işleme parametrelerinin yüzey pürüzlülüğüne etkisinin incelenmesi ve kesme parametrelerinin optimizasyonu

Bu calismada ASTM B574 (Hastelloy C-22) malzemesi uzerinde tornalama islemi sonrasinda kesme parametrelerinin (kesme derinligi, kesme hizi, radius, ilerleme hizi, debi, talas acisi, yanasma acisi) yuzey puruzlugu ve sicaklik uzerindeki etkileri incelenmistir. Kesme parametrelerini etkileyen faktorleri belirlemek icin varyans analizi (ANOVA) uygulanirken, yuzey puruzlulugunu etkileyen parametrelerin optimizasyonu Taguchi ortogonal deney tasarimina dayanan Tepki Yuzeyi Metodolojisi (RSM) ile elde edilmistir. Gelistirilen modellerin yuzey puruzlulugu ve sicaklik tahmini icin gerekli olan dogrulugu oldukca basarilidir. Olcut olarak xa0degerinin kullanildigi calismada ortalama Ra yuzey puruzlugu icin %93.85, dogruluk degeri elde edilmistir. ANOVA analizleri sonucunda %95 guven araliginda, Ra icin en etkili parametreler sirasiyla kesme hizi, yanasma acisi, talas acisi ve debi olarak tespit edilmis ve en dusuk yuzey puruzlugu orani icin en uygun kesme parametre degerleri belirlenmistir.

ACİL SERVİSLER İÇİN BULANIK MANTIK TABANLI BİR KLİNİK KARAR DESTEK SİSTEMİ

Acil servisler, her hastanede olan ve icerisinde ozel birimlerin bulundugu, bircok problemi olan en onemli birimlerinden biridir. Bu sorunlarin basinda, acil servislerin kalabalik olmasi ve acil hasta bakim planlamasinin zorlugu gelmektedir. Bu problemler icin triyaj sistemi gibi uygulamalar kullanilmaktadir. Fakat bu gibi uygulamalarinda problemlere tam olarak cozum getiremedigi bilinmektedir. Bu calismada acil servise gelen hastalarin siniflandirilmasina yonelik bulanik mantik tabanli bir klinik karar destek sistemi (KKDS) gerceklestirilmistir. Calismada Mugla Sitki Kocman Universitesi Egitim ve Arastirma Hastanesinde anonim olmayan 180 hastanin basvuru sikâyetleri ve medikal verileri kullanilmistir. Hastalarin 95i kadin, 85i erkek olup yas ortalamalari 46’dir. Gerceklestirilen sistemin performansini test etmek icin uygulamanin sonuclari ve uzman hekimin kararlari istatistiksel olarak degerlendirilerek (dogruluk, duyarlilik ve ozgulluk) karsilastirilmistir. Sonuc olarak, gerceklestirilen sistemin dogrulugu %83, duyarliligi %87, ozgullugu %76,6 bulunmustur. En son kararin uzman hekime ait olmasi sartiyla bu tur KKDS’nin gelistirilmesi hastanelerin acil servislerinde ozellikle yogun oldugu donemlerde ciddi zaman ve mekân acisindan kazancli olacagi dusunulmektedir.

Büyük ölçekli veri setleri için GPU hızlandırmalı melez bir GA-SVM: Cu-GA-SVM

Bu calismada Genetik Algoritma ve Destek Vektor Makinelerinden olusan melez bir yontemin CUDA tabanli hiz optimizasyonu gerceklestirilmistir. Makine ogrenmesinde, gelistirilen yontemlerin yuksek dogruluk degerlerinde basari vermesi hedeflenir. Ayrica onerilen algoritmanin sonuclari bulurken hizli bir sekilde calismasi da yine hedeflenen bir durumdur. Bu calismada, ozellikle gercek zamanli uygulamalarda onemli bir parametre olan hiz parametresi dikkate alinmakta ve verilerin hizli bir sekilde siniflandirilmasi icin yeni bir GPU teknolojisi kullanilmaktadir. Bunun icin grafik islemciler uzerinde programlama yapmamizi saglayan CUDA programlamadan yararlanilmistir. Siniflandirma algoritmasi olarak genetik algoritmayla optimize edilmis destek vektor makinesi kullanilmistir. Deneyler 384 CUDA cekirdeginden olusan NVIDIA GeForce 940MX ekran kartina sahip bir bilgisayar uzerinde gerceklestirilmistir. Buyuk olcekli veri kumeleri uzerinde yapilan deneylerde, CUDA programlamanin sonuclar uzerinde pozitif etkilerinin oldugu gorulmustur. Bu sekilde makine ogrenmesi uygulamalarinda siniflandirma asamasinda grafik islemciler ile gercek zamanli uygulamalar icin hizli bir sistemin altyapisi olusturulabilir.

Enerji Tasarruflu Bina Tasarımı İçin Isıtma ve Soğutma Yüklerini Regresyon Tabanlı Makine Öğrenmesi Algoritmaları İle Modelleme

Gunumuzde bilisim teknolojileri hemen hemen her alanda kullanilmaktadir. Enerji sektoru de bu alanlardan birisidir. Nufusun gun gectikce artmasiyla birlikte bina sayisi ve binalarin enerji talebi de artmistir. Enerji talebini hafifletmenin bir yolu enerji tasarrufu ozelliklerine sahip verimli binalar tasarlamaktir. Bu calismada sekiz giris degeri (nispi yogunluk, yuzey alani, duvar alani, cati alani, toplam yukseklik, yonlendirme, cam alani ve cam alani dagilimi) ve iki cikis degeri (isitma yuku (HL), sogutma yuku (CL)) olan bir veri setinin, makine ogrenmesi algoritmalari kullanarak analizi yapilmistir. Amac, konutlarin isitma ve sogutma yukunu tahmin edebilen bir model olusturmaktir. Bu parametrelerin dogru bir sekilde tahmin edilmesi, enerji tuketiminin daha iyi kontrol edilmesini kolaylastirmakta ve ayrica, enerji piyasasinda onemli bir sorun olarak gorulen enerji ihtiyacina daha iyi uyan enerji tedarikcisinin seciminde yardimci olmaktadir. Bu kapsamda, veri seti analiz edilirken makine ogrenmesi algoritmalarindan regresyon algoritmalari (Destek Vektor Makinesi (SVM) Regresyonu, Dogrusal Regresyon, Rasgele Orman Regresyonu ve En Yakin Komsu Regresyonu) kullanilmistir. Iki cikis degeri icin sonuclar deneysel olarak her algoritma icin ayri ayri hesaplanmis ve elde edilen sonuclar karsilastirilmistir. Cikan sonuclara gore analiz yaptigimiz veri seti icin, tahmin basarimi acisindan en yakin sonucu bulan algoritma Rastgele Orman Regresyon algoritmasi olmustur.

A computer aided system for calculation of Ki-67 proliferation index

Neuroendocrine tumors (NET) are a heterogeneous group of tumors that can develop in almost every localization of the body. The study is the computation of the Ki-67 proliferation index, which is an important information for physicians, from the NET images. The physician control evaluations were carried out on histopathologic forms taken from healthy and NET-diagnosed patients with the designed automated cell counting system performed. The pathological image analysis of the physician were operated for about 10 minutes while the performed system the average of 5 sec depending on resolution and density of the image. The obtained results were close to 98,71% when compared to the findings of the physician. It has been seen that the realized application has produced very short time and much more accurate in the analysis of these NET images than the analysis of the eyes.

A novel approach for transportation mode detection: Combining t-SNE manifold learning and support vector machines

The aim of this study is to detect transportation modes by using smart phone sensor data. The data are obtained from the GPS, accelerometer and gyroscope sensors of the smartphone. The collected data is divided into 10 second windows and each pattern contains 200 patterns. After the attributes have been determined, the manifold learning algorithm is applied to data set. The obtain features are classified by the Support Vector Machine (SVM) method. In experimental study stage, the performances of three kernel functions of the SVM were compared.