Yazar "Kahramanli, Humar" seçeneğine göre listele
Listeleniyor 1 - 16 / 16
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe Application of ABM to Spectral Features for Emotion Recognition(MEHRAN UNIV ENGINEERING & TECHNOLOGY, 2018) Demircan, Semiye; Kahramanli, HumarER (Emotion Recognition) from speech signals has been among the attractive subjects lately. As known feature extraction and feature selection are most important process steps in ER from speech signals. The aim of present study is to select the most relevant spectral feature subset. The proposed method is based on feature selection with optimization algorithm among the features obtained from speech signals. Firstly, MFCC (Mel-Frequency Cepstrum Coefficients) were extracted from the EmoDB. Several statistical values as maximum, minimum, mean, standard deviation, skewness, kurtosis and median were obtained from MFCC. The next process of study was feature selection which was performed in two stages: In the first stage ABM (Agent-Based Modelling) that is hardly applied to this area was applied to actual features. In the second stageOpt-aiNET optimization algorithm was applied in order to choose the agent group giving the best classification success. The last process of the study is classification. ANN (Artificial Neural Network) and 10 cross-validations were used for classification and evaluation. A narrow comprehension with three emotions was performed in the application. As a result, it was seen that the classification accuracy was rising after applying proposed method. The method was shown promising performance with spectral features.Öğe Application of fuzzy C-means clustering algorithm to spectral features for emotion classification from speech(SPRINGER, 2018) Demircan, Semiye; Kahramanli, HumarIn the present study, emotion recognition from speech signals was performed by using the fuzzy C-means algorithm. Spectral features obtained from speech signals were used as features. The spectral features used were Mel frequency cepstral coefficients and linear prediction coefficients. Certain statistical features were extracted from the spectral features obtained in the study. After the selection of the extracted features, cluster centers were identified by using type-1 fuzzy C-means (FCM) algorithm and used as input to the classifier. Supervised classifiers such as ANN, NB, kNN, and SVM were used for classification. In the study, all seven emotions of the EmoDB database were used. Of the features obtained, FCM clustering was applied to Mel coefficients and obtained clusters centers were used as input for classification. The results showed that using FCM for preprocessing aim increased the success rate. The comparison of the classification methods showed that the maximum success rate was obtained as 92.86% using the SVM classifier.Öğe An Application of Weighted Fuzzy Soft Set Based Decision Making(IEEE, 2013) Allahverdi, Metin; Kahramanli, HumarIn recent years decision making methods have become very important to solve some problems in economics, business, finance and etc. In this paper soft sets and fuzzy soft sets were briefly introduced and then they were applied to hotel choosing problem.Öğe Classification Rule Mining Approach Based on Multiobjective Optimization(IEEE, 2017) Sag, Tahir; Kahramanli, HumarIn this paper, a novel approach for classification rule mining is presented. The remarkable relationship between the rule extraction procedure and the concept of multiobjective optimization is emphasized. The range values of features composing the rules are handled as decision variables in the modelled multiobjective optimization problem. The proposed method is applied to three well-known datasets in literature. These are Iris, Haberman's Survival Data and Pima Indians Diabetes Datasets obtained from machine learning repository of University of California at Irvine (UCI). The classification rules are extracted with 100% accuracy for all datasets. These experimental results are the best outcomes found in literature so far.Öğe Design of a hybrid system for the diabetes and heart diseases(PERGAMON-ELSEVIER SCIENCE LTD, 2008) Kahramanli, Humar; Allahverdi, NovruzData can be classified according to their properties. Classification is implemented by developing a model with existing records by using sample data. One of the aims of classification is to increase the reliability of the results obtained from the data. Fuzzy and crisp values are used together in medical data. Regarding to this, a new method is presented for classification of data of a medical database in this study. Also a hybrid neural network that includes artificial neural network (ANN) and fuzzy neural network (FNN) was developed. Two real-time problem data were investigated for determining the applicability of the proposed method. The data were obtained from the University of California at Irvine (UCI) machine learning repository. The datasets are Pima Indians diabetes and Cleveland heart disease. In order to evaluate the performance of the proposed method accuracy, sensitivity and specificity performance measures that are used commonly in medical classification studies were used. The classification accuracies of these datasets were obtained by k-fold cross-validation. The proposed method achieved accuracy values 84.24% and 86.8% for Pima Indians diabetes dataset and Cleveland heart disease dataset, respectively. It has been observed that these results are one of the best results compared with results obtained from related previous studies and reported in the UCI web sites. (C) 2007 Published by Elsevier Ltd.Öğe Determination of Classification Rules for Heart Diseases(IEEE, 2008) Kahramanli, Humar; Allahverdi, NovruzAlthough Artificial Neural Network (ANN) usually reaches high classification accuracy, the obtained results sometimes may be incomprehensible. This fact is causing a serious problem in data mining applications. The rules that are derived from ANN are needed to be formed to solve this problem and various methods have been improved to extract these rules. In this study for the purpose of extracting rules from ANN which has been trained for classification has been used OptaiNET that is an Artificial Immune Algorithm (AIS) and a set of rules has been formed for heart diseases. Me proposed method is named as OPTBP.Öğe Emotion Recognition via Agent-Based Modelling(IEEE, 2017) Demircan, Semiye; Kahramanli, HumarEmotion recognition is one of the most popular research areas in recent times. Emotion recognition is also made from facial expressions and sound signals, as can be done biomedical signals. Especially when face-to face communication is not possible, emotion can he recognized from the sound data. In this study, emotion recognition was performed from the sound data. One of the most important steps in feeling recognition is feature selection. Feature selection can be done in many different ways. In this study, a new agent-based approach to emotion recognition is presented. The agent-based modeling features were then selected by opt-ainet optimization method. The goal is automatic selection of features that give the best classification accuracy.Öğe EVOLVING RULES FROM NEURAL NETWORKS TRAINED ON BINARY AND CONTINUOUS DATA(NOVA SCIENCE PUBLISHERS, INC, 2010) Kahramanli, Humar; Allahverdi, NovruzAlthough an Artificial Neural Network (ANN) usually reaches high classification accuracy, the obtained results sometimes may be incomprehensible. This fact is causing a serious problem in data mining applications. The rules that are derived from an ANN need to be formed to solve this problem and various methods have been improved to extract these rules. In this study, a new method that uses an Artificial Immune Systems (AIS) algorithm has been presented to extract rules from a trained ANN. The suggested algorithm does not depend on the ANN training algorithms; also, it does not modify the training results. This algorithm takes all input attributes into consideration and extracts rules from a trained neural network efficiently. This study demonstrates the use of AIS algorithms for extracting rules from trained neural networks. The approach consists of three phases: 1. data coding 2. classification of the coded data 3. rule extraction Continuous and noncontinuous values are used together in medical data. Regarding this, two methods are used for data coding and two methods (binary optimisation and real optimisation) are implemented for rule extraction. First, all data are coded binary and the optimal vectors are decoded and used to obtain rules. Then nominal data are coded binary and real data are normalized. After optimization, various intervals for continuous data are obtained and classification accuracy is increased.Öğe Extracting rules for classification problems: AIS based approach(PERGAMON-ELSEVIER SCIENCE LTD, 2009) Kahramanli, Humar; Allahverdi, NovruzAlthough Artificial Neural Network (ANN) usually reaches high classification accuracy, the obtained results in most cases may be incomprehensible. This fact is causing a serious problem in data mining applications. The rules that are derived from ANN are needed to be formed to solve this problem and various methods have been improved to extract these rules. In our previous work, a hybrid neural network was presented for classification (Kahramanli & Allahverdi, 2008). In this study a method that uses Artificial Immune Systems (AIS) algorithm has been presented to extract rules from trained hybrid neural network. The data were obtained from the University of California at Irvine (UCI) machine learning repository. The datasets are Cleveland heart disease and Hepatitis data. The proposed method achieved accuracy values 96.4% and 96.8% for Cleveland heart disease dataset and Hepatitis dataset respectively. It is been observed that these results are one of the best results comparing with results obtained from related previous studies and reported in UCI web sites. (C) 2009 Published by Elsevier Ltd.Öğe Investigation and modeling of the tractive performance of radial tires using off-road vehicles(PERGAMON-ELSEVIER SCIENCE LTD, 2015) Ekinci, Serafettin; Carman, Kazim; Kahramanli, HumarIn order to utilize energy in the most efficient way in off-road vehicles, soil wheel interaction should be investigated carefully since considerable amount of energy is lost due to tractive performance. In this study, the effects of radial tire on tractive performance at three different tire lug heights, axle loads and inflation pressures were experimentally determined. To obtain sufficient performance data, a new single wheel tester was designed and manufactured. Prior to experiments, properties of stubble field were determined. The tractive efficiency was found to increase with increasing dynamic axle load while decreasing with increasing tire inflation pressure. Dynamic axle load of the tire was the major contributory factor in the traction performance as compared with other independent variables. Seven different Artificial Neural Network and two types of Support Vector Regression models have been designed to predict the tractive efficiency. To evaluate the success of system, various statistical measures such as Mean Absolute Error, Root Mean Squared Error and Coefficient Determination have been used. The results show that the Artificial Neural Network model trained using Levenberg Marquardt algorithm has produced more accurate results. (C) 2015 Elsevier Ltd. All rights reserved.Öğe MODELING AND INVESTIGATION OF THE WEAR RESISTANCE OF SALT BATH NITRIDED AISI 4140 VIA ANN(WORLD SCIENTIFIC PUBL CO PTE LTD, 2013) Ekinci, Serafettin; Akdemir, Ahmet; Kahramanli, HumarNitriding is usually used to improve the surface properties of steel materials. In this way, the wear resistance of steels is improved. We conducted a series of studies in order to investigate the microstructural, mechanical and tribological properties of salt bath nitrided AISI 4140 steel. The present study has two parts. For the first phase, the tribological behavior of the AISI 4140 steel which was nitrided in sulfinuz salt bath (SBN) was compared to the behavior of the same steel which was untreated. After surface characterization using metallography, microhardness and sliding wear tests were performed on a block-on-cylinder machine in which carbonized AISI 52100 steel discs were used as the counter face. For the examined AISI 4140 steel samples with and without surface treatment, the evolution of both the friction coefficient and of the wear behavior were determined under various loads, at different sliding velocities and a total sliding distance of 1000 m. The test results showed that wear resistance increased with the nitriding process, friction coefficient decreased due to the sulfur in salt bath and friction coefficient depended systematically on surface hardness. For the second part of this study, four artificial neural network (ANN) models were designed to predict the weight loss and friction coefficient of the nitrided and unnitrided AISI 4140 steel. Load, velocity and sliding distance were used as input. Back-propagation algorithm was chosen for training the ANN. Statistical measurements of R-2, MAE and RMSE were employed to evaluate the success of the systems. The results showed that all the systems produced successful results.Öğe A NEW ACCURATE AND EFFICIENT APPROACH TO EXTRACT CLASSIFICATION RULES(GAZI UNIV, FAC ENGINEERING ARCHITECTURE, 2014) Koklu, Murat; Kahramanli, Humar; Allahverdi, NovruzA new method for extracting rules from multi-class datasets was proposed in this study. The proposed method was applied to 4 different data set. Discrete and real attributes were decoded in different ways. Discrete attributes were coded as binary whereas real attributes were coded by using two real values These values indicate the midpoint and the expansion of intervals of the attributes that form the rules. Classification success was used as fitness function of rule extraction. CLONALG which is Artificial Immune Systems (AIS) algorithm was used to optimize the fitness function. To apply the proposed method Iris, Wine, Glass and Abalone datasets were used. The datasets were obtained from machine learning repository of University of California at Irvine (UCI). The proposed method achieved prediction accuracy ratios of 100%, 99,44%, 77,10%, and 62,59% for Iris, Wine, Glass and Abalone datasets, respectively. When it is compared with the previous studies it has been seen that the proposed method achieved more successful results and has advantage in terms of complexity.Öğe A NEW APPROACH TO CLASSIFICATION RULE EXTRACTION PROBLEM BY THE REAL VALUE CODING(ICIC INTERNATIONAL, 2012) Koklu, Murat; Kahramanli, Humar; Allahverdi, NovruzIn this study a new method that uses artificial immune system (AIS) algorithm has been presented to extract rules from medical related dataset. Four real life problems data were investigated for determining feasibility of the proposed method. The data were obtained from machine learning repository of University of California at Irvine (UCI). The datasets were obtained from Iris Dataset which is the multi-class problem, Pima Indian Diabetes Dataset and two different Wisconsin Breast Cancer datasets. The proposed method achieved prediciton accuracy ratios of 100%, 77.2%, 98.54% and 95.61% for the Iris, Pima Indians Diabetes, Wisconsin Breast Cancer (original) and Wisconsin Breast Cancer (diagnostic) datasets, respectively. It has been observed that these results are better than the results obtained from related previous studies.Öğe Rule extraction from trained adaptive neural networks using artificial immune systems(PERGAMON-ELSEVIER SCIENCE LTD, 2009) Kahramanli, Humar; Allahverdi, NovruzAlthough artificial neural network (ANN) usually reaches high classification accuracy, the obtained results sometimes may be incomprehensible. This fact is causing it serious problem in data mining applications. The rules that are derived from ANN are needed to be formed to solve this problem and various methods have been improved to extract these rules. Activation function is critical as the behavior and performance of an ANN model largely depends oil it. So far there have been limited studies with emphasis oil setting a few free parameters in the neuron activation function. ANN's with such activation function Seem to provide better fitting properties than classical architectures with fixed activation function neurons [Xu, S., & Zhang, M. (2005). Data mining - An adaptive neural network model for financial analysis. In Proceedings of the third international conference on information technology and applications]. In this study a new method that uses artificial immune systems (AIS) algorithm has been presented to extract rules from trained adaptive neural network. Two real time problems data were investigated for determining applicability of the proposed method. The data were obtained from University of California at Irvine (UCI) machine learning repository. The datasets were obtained from Breast Cancer disease and ECG data. The proposed method achieved accuracy values 94.59% and 92.3% for ECG and Breast Cancer dataset, respectively. It has been observed that these results arc one of the best results comparing with results obtained from related previous studies and reported in UCI web sites. (c) 2007 Elsevier Ltd. All rights reserved.Öğe A system for detection of Liver Disorders based on Adaptive Neural Networks and Artificial Immune System(WORLD SCIENTIFIC AND ENGINEERING ACAD AND SOC, 2008) Kahramanli, Humar; Allahverdi, NovruzAlthough Artificial Neural Network (ANN) may achieve high accuracy of classification, the knowledge acquired by them is incomprehensible for humans. This fact is causing a serious problem in data mining applications. The rules that are derived from ANN are needed to be formed to solve this problem and various methods have been improved to extract these rules. Selection of the activation function is important in the performance of ANN. Networks with adaptive activation function seem to provide better fitting properties than classical architectures with fixed activation function neurons [1]. In this study, first neural network has been trained with adaptive activation function. Then for the purpose of extracting rules from adaptive ANN which has been trained for classification, OptaiNET that is an Artificial Immune Algorithm (AIS) has been used and a set of rules has been formed for liver disorder.Öğe A Takagi-Sugeno type neuro-fuzzy network for determining child anemia(PERGAMON-ELSEVIER SCIENCE LTD, 2011) Allahverdi, Novruz; Tunali, Ayfer; Isik, Hakan; Kahramanli, HumarDecision-making is a difficult and quite responsible task for doctors. Some of the computer decision models assisted the doctor with some computer decision models. In this study, neuro-fuzzy network has been designed to determine anemia level of a child. The performance analyses have been obtained by leaving-one-out cross-validation. After statistical measurements, it was found that MPE = 0.0018, MAE = 0.2090, MAPE = 0.0511, RMSE = 0.2743 and R-2 = 0.9957 of this developed system. According to these results, the designed neuro-fuzzy network may be considered as adequate close to traditional decision-making methods and thus the designed network can be used effectively for child anemia prediction. (c) 2010 Elsevier Ltd. All rights reserved.
