Bekir Karlik

A novel approach for classification of ECG arrhythmias: Type-2 fuzzy clustering neural network

This paper presents an improved classifier for automated diagnostic systems of electrocardiogram (ECG) arrhythmias. This diagnostic system consists of a combined Fuzzy Clustering Neural Network Algorithm for Classification of ECG Arrhythmias using type-2 fuzzy c-means clustering (T2FCM) algorithm and neural network. Type-2 fuzzy c-means clustering is used to improve performance of neural network. The aim of improving classifiers performance is to constitute the best classification system with high accuracy rate for ECG beats. Ten types of ECG arrhythmias (normal beat, sinus bradycardia, ventricular tachycardia, sinus arrhythmia, atrial premature contraction, paced beat, right bundle branch block, left bundle branch block, atrial fibrillation and atrial flutter) obtained from MIT-BIH database were analyzed. However, the presented structure was tested by experimental ECG records of 92 patients (40 male and 52 female, average age is 39.75+/-19.06). The classification accuracy of an improved classifier in training and testing, namely Type-2 Fuzzy Clustering Neural Network (T2FCNN), was compared with neural network (NN) and fuzzy clustering neural network (FCNN). In T2FCNN architecture, decision making has two stages: forming of the new training set obtained by selection of the best arrhythmia for each arrhythmia class using T2FCM and classification using neural network trained on the new training set. The results are demonstrated that the proposed diagnostic systems achieved high (99%) accuracy rate.

Integration of type-2 fuzzy clustering and wavelet transform in a neural network based ECG classifier

This paper presents a new automated diagnostic system to classification of electrocardiogram (ECG) arrhythmias. The diagnostic system is executed using type-2 fuzzy c-means clustering (T2FCM) algorithm, wavelet transform (WT) and neural network. Method of combining T2FCM and WT is used to improve performance of neural network. We aimed high accuracy rate to classification of ECG beats and constituted the automated diagnostic system to improve of classifiers performance. Ten types of ECG beats selected from MIT-BIH database were used to train the system. Then, this system was tested by the ECG signals of patients. The classification accuracy of the proposed classifier, type-2 fuzzy clustering wavelet neural network (T2FCWNN), is compared with the structures formed by type-1 FCM and WT. Process of T2FCWNN architecture is realized on three stages. First stage is formed the new training set obtained by selection of the best segments for each arrhythmia class using T2FCM. Second stage is feature extraction by WT on the new training set. Third stage is classification of the extracted features using neural network. The research showed that accuracy rate was found as 99% using this system.

Diagnosing diabetes using neural networks on small mobile devices

Pervasive computing is often mentioned in the context of improving healthcare. This paper presents a novel approach for diagnosing diabetes using neural networks and pervasive healthcare computing technologies. The recent developments in small mobile devices and wireless communications provide a strong motivation to develop new software techniques and mobile services for pervasive healthcare computing. A distributed end-to-end pervasive healthcare system utilizing neural network computations for diagnosing illnesses was developed. This work presents the initial results for a simple client (patients PDA) and server (powerful desktop PC) two-tier pervasive healthcare architecture. The computations of neural network operations on both client and server sides and wireless network communications between them are optimized for real time use of pervasive healthcare services.

An artificial neural networks approach on automobile pricing

The aim of this study is to find an automobile pricing model using artificial neural networks (ANN). As commonly known, pricing is a difficult matter for both automobile manufacturers and buyers/sellers. Developing a neural networks based on the technical properties of automobiles will allow both groups to price autos with great ease. However, in this study there are two basic assumptions. The first is that supply and demand are in equilibrium and they have no positive or negative effect on pricing. Alfred Marshall [Alfred Marshall. (1920). Principles of economics (Vol. 9). Macmillan] describes how the price and availability of goods and services are related to consumer demand in competitive markets in the Law of supply and demand. The second is that our data set represents the whole market since we will determine market prices of other automobiles according to the network that is trained by this dataset. Proposed novel model estimates prices of automobiles on a stable market from their technical and physical properties.

A method for removing low varying frequency trend from ECG signal

The aim of this work is based on recovering the ECG signal from low frequency varying noise. The method used for this purpose is simple, the well known moving average filter. For this work the moving average filter is used to operate as a highpass filter. The filtering system functions basically to remove the trend added to the ECG signal. The authors have seen that the moving average filter recovers perfectly the low varying trend. Then by subtracting the recorded data from the trend, the noise-free ECG signal is obtained. Moreover, the effectiveness of the proposed method is shown in the frequency domain.

Automatic recognition of retinopathy diseases by using wavelet based neural network

In this study, recognition of five types of retina disorders and normal retina has been studied. The names of these five different Retinopathies are: Diabetic Retinopathy, Hypertensive retinopathy, Macular Degeneration, Vein Branch Oclusion, Vitreus hemorrhage, and normal retina. A wavelet based neural network architecture has been used to diagnose retinopathy automatically. In the process, the retina images were pre-processed and resized. Later, feature extraction has been done before applying into classifier. The performance of proposed method has been found very high. The recognition rates were found %50, %70, %83, %90, %93 and %95 for testing five retinopathy cases respectively.

Hazardous Odor Recognition by CMAC Based Neural Networks

Electronic noses are being developed as systems for the automated detection and classification of odors, vapors, and gases. Artificial neural networks (ANNs) have been used to analyze complex data and to recognize patterns, and have shown promising results in recognition of volatile compounds and odors in electronic nose applications. When an ANN is combined with a sensor array, the number of detectable chemicals is generally greater than the number of unique sensor types. The odor sensing system should be extended to new areas since its standard style where the output pattern from multiple sensors with partially overlapped specificity is recognized by a neural network or multivariate analysis. This paper describes the design, implementation and performance evaluations of the application developed for hazardous odor recognition using Cerebellar Model Articulation Controller (CMAC) based neural networks.

Comparison neural networks models for short term forecasting of natural gas consumption in Istanbul

The aim of this study is to find a suitable natural gas energy forecasting model for daily and weekly values of Istanbul by using artificial neural networks(ANN). As it is known, accurate forecasting is important for both gas distributors and consumers. On the view point of distributors, with accurate forecasting the number of false alarms would be significantly decreased and trans ship limits would be scheduled. On the view point of consumers, there will be no disconnect and breakdown etc. In this study, a wide factor analyzing is done in order to find the factors that effect the gas consumptions. Found results were applied to ANN feed forward back propagation algorithms. The reasons behind choosing ANN are the ability of forecasting future values of more than one variable at the same time and to model the nonlinear relation in the data structure. Performance comparisons of seven different algorithms were done.

Neural network methodology for heat transfer enhancement data

Purpose – The purpose of this paper is to study experimentally enhancement of heat transfer in a tube with axial swirling‐flow promoters. The geometric features of flow geometry to improve heat transfer can be selected in order to yield the maximum opposite reduction in heat exchange flow irreversibility by using exergy‐destruction method. The paper seeks to illustrate the use of neural network approach to analyze heat transfer enhancement data for further study in the scope of the experimental program.Design/methodology/approach – For this purpose, 402 experimental measurements are collected. About 225 of those are used as training data for neural networks, the rest is used for testing. Then, these testing results of artificial neural network (ANN) and experimental data are compared. A formula for presenting exergy loses in a tubular heat exchanger is derived first and then the thermodynamic optimum instead of economic optimum is found by minimizing the exergy losses in the system.Findings – Results from...

A novel mobile epilepsy warning system

This paper presents a new design of mobile epilepsy warning system for medical application in telemedical environment. Mobile Epilepsy Warning System (MEWS) consists of a wig with a cap equipped with sensors to get Electroencephalogram (EEG) signals, a collector which is used for converting signals to data, Global Positioning System (GPS), a Personal Digital Assistant (PDA) which has Global System for Mobile (GSM) module and execute Artificial Neural Network (ANN) software to test current patient EEG data with pre-learned data, and a calling center for patient assistance or support. The system works as individual sensors obtain EEG signals from patient who has epilepsy and establishes a communication between the patient and Calling Center (CC) in case of an epileptic attack. MEWS learning process has artificial neural network classifier, which consists of Multi Layered Perceptron (MLP) neural networks structure and back-propagation training algorithm.