This study concentrates on the comparison of the discriminant functions and the decision tree ind... more This study concentrates on the comparison of the discriminant functions and the decision tree induction techniques in antepartum fetal evaluation. These classification techniques are applied to antenatal fetal risk assessment problem and the performances, the computational complexities and the importance of each technique in terms of diagnostic clues are observed. The task is to investigate the Doppler ultrasound measurements of umbilical artery (UA) to relate the health conditions of fetuses using discriminant functions ...
2007 4th IEEE Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, 2007
ABSTRACT Reliable arrhythmia classification from complex electrocardiogram (ECG) signals is one o... more ABSTRACT Reliable arrhythmia classification from complex electrocardiogram (ECG) signals is one of the most challenging pattern recognition problems. Several individual classifiers have been studied in the ECG domain. Also, parallel and serial classifier fusion systems have been proposed to increase the reliability. In this study, we are mainly interested in producing high confident arrhythmia classification results to be applicable in diagnostic decision support systems. We first experiment and compare two common techniques: support vector machines (SVM) and logistic regression (LR). Then, we propose a two- stage serial fusion classifier system based on SVM's rejection option. We relate the SVM's distance outputs to confidence measure and reject to classify ambiguous samples with first level SVM classifier. A non-symmetric thresholding scheme is applied: two different rejection distance thresholds have been defined for positive and negative ECG samples. The rejected samples have been forwarded to a second stage LR classifier. Finally we choose a way to combine the classifiers decisions to obtain a final decision rule. The experiments have been performed on UCI Arrhythmia Database.
Istanbul Stock Exchange (ISE) stock market is not an efficient market. In this paper, we show how... more Istanbul Stock Exchange (ISE) stock market is not an efficient market. In this paper, we show how localized and global artificial neural network (ANN) models are used for risk estimation of asset returns. Mixture of Experts (MoE) and Recurrent Neural Networks (RNN) are more powerful to say that Efficient Market Hypothesis (EMH) is violated. ISE index XU100 is studied using daily data over a 14-year period using MoE and RNN neural networks and also Glosten-Jaganathan-Runkle (GJR) volatility models. The results suggest that localized neural approaches have the strength in modeling the risk in stock market time series data set of XU100.
Proceedings of 1993 International Conference on Neural Networks (IJCNN-93-Nagoya, Japan), 1993
Abstract The comparative performances of distributed and local neural networks for the speech rec... more Abstract The comparative performances of distributed and local neural networks for the speech recognition problem is investigated. Distributed networks' hidden units use the signoid nonlinearity with global response. We have used the backpropagation rule with three error measures: mean square error, cross entropy, and combinational performance. The hidden units of local networks respond only to inputs in a certain local region in the input space. We used k-nearest neighbor (kNN), Gaussian-based kNN, learning vector ...
This study proposes an intelligent data analysis approach to investigate and interpret the distin... more This study proposes an intelligent data analysis approach to investigate and interpret the distinctive factors of diabetes mellitus patients with and without ischemic (non-embolic type) stroke in a small population. The database consists of a total of 16 features collected from 44 diabetic patients. Features include age, gender, duration of diabetes, cholesterol, high density lipoprotein, triglyceride levels, neuropathy, nephropathy, retinopathy, peripheral vascular disease, myocardial infarction rate, glucose level, medication and blood pressure. Metric and non-metric features are distinguished. First, the mean and covariance of the data are estimated and the correlated components are observed. Second, major components are extracted by principal component analysis. Finally, as common examples of local and global classification approach, a k-nearest neighbor and a high-degree polynomial classifier such as multilayer perceptron are employed for classification with all the components ...
Proceedings of the IEEE 12th Signal Processing and Communications Applications Conference, 2004., 2004
In this study, the objective is to develop a new combined method for efficient compression of the... more In this study, the objective is to develop a new combined method for efficient compression of the classical 3D object mesh representation. This can be realized in two primary steps: mesh connectivity coding and data (geometry) compression. For realizing the first step, the algorithm of Isenburg (2002) has been employed. For the second step, vector quantization methods have been used
The representation of large document databases, consisting of Web pages, articles and book and ma... more The representation of large document databases, consisting of Web pages, articles and book and magazine titles, in terms of matrices for the purpose of text querying and retrieval simplifies and expedites the querying process. In the literature, dimensionality reduction techniques based on singular value decomposition and principal component analysis have been proposed to reduce the high computational complexity resulting from
In some machine learning problems, the dataset has multiple views which may be obtained using dif... more In some machine learning problems, the dataset has multiple views which may be obtained using different sensors or applying different sampling techniques. These views may have sufficient or partial information about the target concept. In this paper, a method that we called parallel interacting multiview learning (PIML) is proposed in which the views interact during the training process using the
Artificial Intelligence and Applications / 718: Modelling, Identification, and Control, 2011
Page 1. ISTANBUL STOCK EXCHANGE (ISE) 100 PREDICTION USING SUPPORT VECTOR PREDICTORS (SVP) M.Serd... more Page 1. ISTANBUL STOCK EXCHANGE (ISE) 100 PREDICTION USING SUPPORT VECTOR PREDICTORS (SVP) M.Serdar Yümlü Department of Computer Engineering Boğaziçi University 34342 Bebek, Istanbul, Turkey yumlu2@yahoo.com ...
Location area (LA) management is a very important problem in mobile networks. In general, registr... more Location area (LA) management is a very important problem in mobile networks. In general, registration and paging costs are associated with tracking the current location of a mobile user. Considering minimizing the total of paging and registration costs as the main objective, the aim is to provide corresponding cell-to-switch and cell-to-LA assignments. This paper compares three well-known evolutionary algorithms to
This study presents parallel implementation of a vector quantization (VQ) based text-independent ... more This study presents parallel implementation of a vector quantization (VQ) based text-independent speaker identification system that uses Melfrequency cepstrum coefficients (MFCC) for feature extraction, Linde-Buzo-Gray (LBG) VQ algorithm for pattern matching ...
This study concentrates on the comparison of the discriminant functions and the decision tree ind... more This study concentrates on the comparison of the discriminant functions and the decision tree induction techniques in antepartum fetal evaluation. These classification techniques are applied to antenatal fetal risk assessment problem and the performances, the computational complexities and the importance of each technique in terms of diagnostic clues are observed. The task is to investigate the Doppler ultrasound measurements of umbilical artery (UA) to relate the health conditions of fetuses using discriminant functions ...
2007 4th IEEE Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, 2007
ABSTRACT Reliable arrhythmia classification from complex electrocardiogram (ECG) signals is one o... more ABSTRACT Reliable arrhythmia classification from complex electrocardiogram (ECG) signals is one of the most challenging pattern recognition problems. Several individual classifiers have been studied in the ECG domain. Also, parallel and serial classifier fusion systems have been proposed to increase the reliability. In this study, we are mainly interested in producing high confident arrhythmia classification results to be applicable in diagnostic decision support systems. We first experiment and compare two common techniques: support vector machines (SVM) and logistic regression (LR). Then, we propose a two- stage serial fusion classifier system based on SVM's rejection option. We relate the SVM's distance outputs to confidence measure and reject to classify ambiguous samples with first level SVM classifier. A non-symmetric thresholding scheme is applied: two different rejection distance thresholds have been defined for positive and negative ECG samples. The rejected samples have been forwarded to a second stage LR classifier. Finally we choose a way to combine the classifiers decisions to obtain a final decision rule. The experiments have been performed on UCI Arrhythmia Database.
Istanbul Stock Exchange (ISE) stock market is not an efficient market. In this paper, we show how... more Istanbul Stock Exchange (ISE) stock market is not an efficient market. In this paper, we show how localized and global artificial neural network (ANN) models are used for risk estimation of asset returns. Mixture of Experts (MoE) and Recurrent Neural Networks (RNN) are more powerful to say that Efficient Market Hypothesis (EMH) is violated. ISE index XU100 is studied using daily data over a 14-year period using MoE and RNN neural networks and also Glosten-Jaganathan-Runkle (GJR) volatility models. The results suggest that localized neural approaches have the strength in modeling the risk in stock market time series data set of XU100.
Proceedings of 1993 International Conference on Neural Networks (IJCNN-93-Nagoya, Japan), 1993
Abstract The comparative performances of distributed and local neural networks for the speech rec... more Abstract The comparative performances of distributed and local neural networks for the speech recognition problem is investigated. Distributed networks' hidden units use the signoid nonlinearity with global response. We have used the backpropagation rule with three error measures: mean square error, cross entropy, and combinational performance. The hidden units of local networks respond only to inputs in a certain local region in the input space. We used k-nearest neighbor (kNN), Gaussian-based kNN, learning vector ...
This study proposes an intelligent data analysis approach to investigate and interpret the distin... more This study proposes an intelligent data analysis approach to investigate and interpret the distinctive factors of diabetes mellitus patients with and without ischemic (non-embolic type) stroke in a small population. The database consists of a total of 16 features collected from 44 diabetic patients. Features include age, gender, duration of diabetes, cholesterol, high density lipoprotein, triglyceride levels, neuropathy, nephropathy, retinopathy, peripheral vascular disease, myocardial infarction rate, glucose level, medication and blood pressure. Metric and non-metric features are distinguished. First, the mean and covariance of the data are estimated and the correlated components are observed. Second, major components are extracted by principal component analysis. Finally, as common examples of local and global classification approach, a k-nearest neighbor and a high-degree polynomial classifier such as multilayer perceptron are employed for classification with all the components ...
Proceedings of the IEEE 12th Signal Processing and Communications Applications Conference, 2004., 2004
In this study, the objective is to develop a new combined method for efficient compression of the... more In this study, the objective is to develop a new combined method for efficient compression of the classical 3D object mesh representation. This can be realized in two primary steps: mesh connectivity coding and data (geometry) compression. For realizing the first step, the algorithm of Isenburg (2002) has been employed. For the second step, vector quantization methods have been used
The representation of large document databases, consisting of Web pages, articles and book and ma... more The representation of large document databases, consisting of Web pages, articles and book and magazine titles, in terms of matrices for the purpose of text querying and retrieval simplifies and expedites the querying process. In the literature, dimensionality reduction techniques based on singular value decomposition and principal component analysis have been proposed to reduce the high computational complexity resulting from
In some machine learning problems, the dataset has multiple views which may be obtained using dif... more In some machine learning problems, the dataset has multiple views which may be obtained using different sensors or applying different sampling techniques. These views may have sufficient or partial information about the target concept. In this paper, a method that we called parallel interacting multiview learning (PIML) is proposed in which the views interact during the training process using the
Artificial Intelligence and Applications / 718: Modelling, Identification, and Control, 2011
Page 1. ISTANBUL STOCK EXCHANGE (ISE) 100 PREDICTION USING SUPPORT VECTOR PREDICTORS (SVP) M.Serd... more Page 1. ISTANBUL STOCK EXCHANGE (ISE) 100 PREDICTION USING SUPPORT VECTOR PREDICTORS (SVP) M.Serdar Yümlü Department of Computer Engineering Boğaziçi University 34342 Bebek, Istanbul, Turkey yumlu2@yahoo.com ...
Location area (LA) management is a very important problem in mobile networks. In general, registr... more Location area (LA) management is a very important problem in mobile networks. In general, registration and paging costs are associated with tracking the current location of a mobile user. Considering minimizing the total of paging and registration costs as the main objective, the aim is to provide corresponding cell-to-switch and cell-to-LA assignments. This paper compares three well-known evolutionary algorithms to
This study presents parallel implementation of a vector quantization (VQ) based text-independent ... more This study presents parallel implementation of a vector quantization (VQ) based text-independent speaker identification system that uses Melfrequency cepstrum coefficients (MFCC) for feature extraction, Linde-Buzo-Gray (LBG) VQ algorithm for pattern matching ...
Uploads
Papers by F. Gurgen