Gulay Tezel

A new method for classification of ECG arrhythmias using neural network with adaptive activation function

In this study, new neural network models with adaptive activation function (NNAAF) were implemented to classify ECG arrhythmias. Our NNAAF models included three types named as NNAAF-1, NNAAF-2 and NNAAf-3. Activation functions with adjustable free parameters were used in hidden neurons of these models to improve classical MLP network. In addition, these three NNAAF models were compared with the MLP model implemented in similar conditions. Ten different types of ECG arrhythmias were selected from MIT-BIH ECG Arrhythmias Database to train NNAAFs and MLP models. Moreover, all models tested by the ECG signals of 92 patients (40 males and 52 females, average age is 39.75+/-19.06). The average accuracy rate of all models in the training processing was found as 99.92%. The average accuracy rate of the all models in the test phases was obtained as 98.19.

A genetic algorithm-support vector machine method with parameter optimization for selecting the tag SNPs

SNPs (Single Nucleotide Polymorphisms) include millions of changes in human genome, and therefore, are promising tools for disease-gene association studies. However, this kind of studies is constrained by the high expense of genotyping millions of SNPs. For this reason, it is required to obtain a suitable subset of SNPs to accurately represent the rest of SNPs. For this purpose, many methods have been developed to select a convenient subset of tag SNPs, but all of them only provide low prediction accuracy. In the present study, a brand new method is developed and introduced as GA-SVM with parameter optimization. This method benefits from support vector machine (SVM) and genetic algorithm (GA) to predict SNPs and to select tag SNPs, respectively. Furthermore, it also uses particle swarm optimization (PSO) algorithm to optimize C and γ parameters of support vector machine. It is experimentally tested on a wide range of datasets, and the obtained results demonstrate that this method can provide better prediction accuracy in identifying tag SNPs compared to other methods at present.

Estimation of the Change in Lake Water Level by Artificial Intelligence Methods

In this study, five different artificial intelligence methods, including Artificial Neural Networks based on Particle Swarm Optimization (PSO-ANN), Support Vector Regression (SVR), Multi- Layer Artificial Neural Networks (MLP), Radial Basis Neural Networks (RBNN) and Adaptive Network Based Fuzzy Inference System (ANFIS), were used to estimate monthly water level change in Lake Beysehir. By using different input combinations consisting of monthly Inflow - Lost flow (I), Precipitation (P), Evaporation (E) and Outflow (O), efforts were made to estimate the change in water level (L). Performance of models established was evaluated using root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE) and coefficient of determination (R2). According to the results of models, ε-SVR model was obtained as the most successful model to estimate monthly water level of Lake Beysehir.

Monthly evaporation forecasting using artificial neural networks and support vector machines

Evaporation is one of the most important components of the hydrological cycle, but is relatively difficult to estimate, due to its complexity, as it can be influenced by numerous factors. Estimation of evaporation is important for the design of reservoirs, especially in arid and semi-arid areas. Artificial neural network methods and support vector machines (SVM) are frequently utilized to estimate evaporation and other hydrological variables. In this study, usability of artificial neural networks (ANNs) (multilayer perceptron (MLP) and radial basis function network (RBFN)) and ε-support vector regression (SVR) artificial intelligence methods was investigated to estimate monthly pan evaporation. For this aim, temperature, relative humidity, wind speed, and precipitation data for the period 1972 to 2005 from Beysehir meteorology station were used as input variables while pan evaporation values were used as output. The Romanenko and Meyer method was also considered for the comparison. The results were compared with observed class A pan evaporation data. In MLP method, four different training algorithms, gradient descent with momentum and adaptive learning rule backpropagation (GDX), Levenberg–Marquardt (LVM), scaled conjugate gradient (SCG), and resilient backpropagation (RBP), were used. Also, ε-SVR model was used as SVR model. The models were designed via 10-fold cross-validation (CV); algorithm performance was assessed via mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). According to the performance criteria, the ANN algorithms and ε-SVR had similar results. The ANNs and ε-SVR methods were found to perform better than the Romanenko and Meyer methods. Consequently, the best performance using the test data was obtained using SCG(4,2,2,1) with R2 = 0.905.

How to Select Tag SNPs in Genetic Association Studies? The CLONTagger Method with Parameter Optimization

Selection of genetic variants is a crucial first step in the rational design of studies aimed at explaining individual differences in susceptibility to complex human diseases or health intervention outcomes; for example, in the emerging fields of pharmacogenomics, nutrigenomics, and vaccinomics. While single nucleotide polymorphisms (SNPs) are frequently employed in these studies, the cost of genotyping a huge number of SNPs remains a limiting factor, particularly in low and middle income countries. Therefore, it is important to detect a subset of SNPs to represent the rest of SNPs with maximum possible accuracy. The present study introduces a new method, CLONTagger with parameter optimization, which uses Support Vector Machine (SVM) to predict the rest of SNPs and Clonal Selection Algorithm (CLONALG) to select tag SNPs. Furthermore, the Particle Swarm Optimization algorithm is preferred for the optimization of C and γ parameters of the Support Vector Machine. Additionally, using many datasets, we compared the proposed new method with the tag SNP selection algorithms present in literature. Our results suggest that the CLONTagger with parameter optimization can identify tag SNPs with better prediction accuracy than other methods. Application-oriented studies are warranted to evaluate the utility of this method in future research in human genetics and study of the genetic components of variable responses to drugs, nutrition, and vaccines.

A new neural network with adaptive activation function for classification of ECG arrhythmias

This study presents a comparative study of the classification accuracy of ECG signals using a well-known neural network architecture named multilayered perceptron (MLP) with backpropagation training algorithm, and a new neural network with adaptive activation function (AAFNN) for classification of ECG arrhythmias. The ECG signals are taken from MIT-BIH ECG database, which are used to classify ten different arrhythmias for training. These are normal sinus rhythm, sinus bradycardia, ventricular tachycardia, sinus arrhythmia, atrial premature contraction, paced beat, right bundle branch block, left bundle branch block, atrial fibrillation and atrial flutter. For testing, the proposed structures were trained by backpropagation algorithm. Both of them tested using experimental ECG records of 10 patients (7 male and 3 female, average age is 33.8±16.4). The results show that neural network with adaptive activation function is more suitable for biomedical data like as ECG in the classification problems and training speed is much faster than neural network with fixed sigmoid activation function.

Automatic sleep staging based on SVD, VMD, HHT and morphological features of single-lead ECG signal

Abstract Electroencephalogram (EEG) signals, which are among the primary polysomnography (PSG) signals used for sleep staging, are difficult to obtain and interpret. However, it is much easier to obtain and interpret electrocardiogram (ECG) signals. The use of ECG signals for automatic sleep staging systems could bring practicality to these systems. In this study, ECG signals were used to identify the wake (W), non-rapid eye movement (NREM) and rapid eye movement (REM) stages of the sleep data from two different databases with 17,758 epochs of 28 subjects (21 healthy subjects and 7 obstructive sleep apnea (OSA) patients) in total. Four different methods were used to extract features from these signals: Singular Value Decomposition (SVD), Variational Mode Decomposition (VMD), Hilbert Huang Transform (HHT), and Morphological method. As a result of applying the methods separately, four different data sets were obtained. The four different datasets were given to the Wrapper Subset Evaluation system with the best-first search algorithm. After the feature selection procedure, the datasets were separately classified by using the Random Forest classifier. The results were interpreted by using the essential statistical criteria. Among the methods, morphological method was the most successful and it was followed by SVD in terms of success rate for both two databases. For the first database, the mean classification accuracy rate, Kappa coefficient and mean F-measure value of the Morphological method were found as 87.11%, 0.7369, 0.869 for the healthy and 78.08%, 0.5715, 0.782 for the patient, respectively. For the second database, the same statistical measures were determined as 77.02%, 0.4308, 0.755 for the healthy and 76.79%, 0.5227, 0.759 for the patient, respectively. The performance results of the study, which is consistent with real life applications, were compared with the previous studies in this field listed in the literature.

Tag SNP selection using similarity associations between SNPs

Genetic changes that may be associated with complex diseases are tried to be determined by means of many genome-wide association studies. Single Nucleotide Polymorphisms (SNPs) are used primarily in these studies since they comprise a large part of these genetic changes. Statistical importance of the genome-wide association study is directly related to the number of individuals and SNPs. However, it is still very costly and time-consuming to genotype all SNPs inside the candidate area for many individuals in very large-scale association studies. For this reason, with a small error, it is necessary to select an appropriate subset of all SNPs that will represent the rest of SNPs. These selected SNPs are called tag SNPs or haplotype tag SNPs (tag SNPs or htSNPs). It is essential in tag SNP selection to determine minimum tag SNP set with very good prediction accuracy. In this study, while Clonal Selection Algorithm (CLONALG) was used as tag SNP selection method, a new method named CLONSim, in which similarity association between SNPs was used as the prediction method for the rest of SNPs was proposed. The proposed method was compared with BPSO (Binary Particle Swarm Optimization) and CLONTagger methods with parameter optimization using datasets of different sizes. Experiment results showed that the proposed method could identify tag SNPs significantly faster.

Effect of the Hilbert-Huang transform method on sleep staging

Sleep scoring is performed by examining the recorded electroencephalogram (EEG) and some other signals recorded by a polysomnography (PSG) device. This process is considered more reliable as it is done manually by experts. However, due to the fact that experts may also be mistaken, it has led to an increase in the importance given to automatic sleep staging studies. Many methods have been tested on the signals in order to increase the success of these systems. In this study, an automatic sleep staging system was implemented using the Hilbert-Huang transformation method which is a new time-frequency analysis type. In the study, EEG signals from 5 subjects were used in the sleep laboratory. In the 5-class (Alpha, Beta, Theta, Delta and Spindle bands) applications, the highest classification success was 84.75% as a result of sequential feature selection method.

Elimination of EMG artifacts from EEG signal in sleep staging

Sleep staging is a tiring and time-consuming process for the experts. Thus, attention given to automatic sleep staging studies is increasing gradually. Many factors such as effects of EOG and EKG signals to EEG result in contaminated signals rather than clear recorded signals. EMG contamination to EEG is among that kind of factors. In this study, some filters and Discrete Wavelet Transform based EMG artifact elimination process were evaluated on the performance of sleep staging process. Features were extracted from cleaned EEG signals and subjected to a classifier to conduct sleep staging. By using test classification accuracy as a measure of performance, the method giving highest accuracy was tried to be found. Artificial Neural Networks was used in the applications and Discrete Wavelet Transform was found to be the method giving the highest accuracy.