Search

In this paper we deal with the problem of measuring the similarity between training and tests datasets in the context of transfer learning (TL) for image classification. TL tries to transfer knowledge from a source domain, where labelled training samples are abundant but the data may follow a different distribution, to a target domain, where labelled training samples are scarce or even unavailable, assuming that the domains are related. Thus, the requirements w.r.t. the availability of labelled training samples in the target domain are reduced. In particular, if no labelled target data are available, it is inherently difficult to find a robust measure of relatedness between the source and target domains. This is of crucial importance for the performance of TL, because the knowledge transfer between unrelated data may lead to negative transfer, i.e. to a decrease of classification performance after transfer. We address the problem of measuring the relatedness between source and target datasets and investigate three different strategies to predict and, consequently, to avoid negative transfer in this paper. The first strategy is based on circular validation. The second strategy relies on the Maximum Mean Discrepancy (MMD) similarity metric, whereas the third one is an extension of MMD which incorporates the knowledge about the class labels in the source domain. Our method is evaluated using two different benchmark datasets. The experiments highlight the strengths and weaknesses of the investigated methods. We also show that it is possible to reduce the amount of negative transfer using these strategies for a TL method and to generate a consistent performance improvement over the whole dataset.

Recently, there has been an increasing trend to propose computer aided diagnosis systems for biomedical pattern recognition. A computer aided diagnosis method, which aims higher classification accuracy, is developed to classify the biomedical dataset. This process includes two types of machine learning algorithms: feature selection and classification. In this method, firstly, features were extracted from biomedical dataset, then the extracted features were classified by hybrid AdaBoost-Support Vector Machines (SVM) classifier structure. For feature selection, Forward Feature Selection (FFS) and Principal Component Analysis (PCA) algorithms were used, and the performance of the feature selection algorithms was tested by AdaBoost-SVM classifier. Following it, advantages and disadvantages of these algorithms were evaluated. Wisconsin Breast Cancer (WBC), Pima Diabetes (PD), Heart (Statlog) biomedical datasets were taken from UCI database and Electrocardiogram (ECG) signals were taken from Physionet ECG Database, and were used to test the proposed hybrid structure. The used two hybrid structures and other studies in the literature were compared with our findings. The obtained results show that the proposed hybrid structure has high classification accuracy for biomedical data classification

The spam detection is a big issue in mobile message communication due to which mobile message communication is insecure. In order to tackle this problem, an accurate and precise method is needed to detect the spam in mobile message communication. We proposed the applications of the machine learning-based spam detection method for accurate detection. In this technique, machine learning classifiers such as Logistic regression (LR), K-nearest neighbor (K-NN), and decision tree (DT) are used for classification of ham and spam messages in mobile device communication. The SMS spam collection data set is used for testing the method. The dataset is split into two categories for training and testing the research. The results of the experiments demonstrated that the classification performance of LR is high as compared with K-NN and DT, and the LR achieved a high accuracy of 99%. Additionally, the proposed method performance is good as compared with the existing state-of-the-art methods. Ministry of Education of the People's Republic of China [DCA190332].

Chronic obstructive pulmonary disease (COPD) is one of the deadliest diseases which cannot be treated but can be kept under control in certain stages. COPD has five severities, including at-risk, mild, moderate, severe, and very severe stages. Diagnosis of COPD at early stages needs additional clinical tests for even experienced specialists. The study aims at detecting the severity of the COPD to start treatment for preventing the progression of the disease to the next levels. We analyzed 12-channel lung sounds with different COPD severities from RespiratoryDatabase@TR. The lung sounds were recorded from the clinical auscultation points from 41 patients on posterior (chest) and anterior (back) sides. 3D second-order difference plot was applied to extract characteristic abnormalities on lung sounds. Cuboid and octant-based quantizations were utilized to extract characteristic abnormalities on chaos plot. Deep extreme learning machines classifier (deep ELM), which is one of the most stable and fast deep learning algorithms, was utilized in the classification stage. Novel HessELM and LuELM autoencoder kernels were adapted to deep ELM and reached higher generalization capabilities with a faster training speed against the conventional ELM autoencoder. The proposed deep ELM model with LuELM autoecoder has separated five COPD severities with classification performance rates of 94.31%, 94.28%, 98.76%, and 0.9659 for overall accuracy, weighted-sensitivity, weighted-specificity, and area under the curve (AUC) value, respectively. The proposed deep analysis of 12-channel lung sounds provides a standardized and entire lung assessment for identification of COPD severity. Our study is a pioneering approach that directly focuses on lung sounds. Novel deep ELM kernels have performed a higher generalization and fast training compared to conventional kernels.

In this study we analyze the signal classification performances of various classifiers for deterministic signals under the additive White Gaussian Noise (WGN) in a wide range of signal to noise ratio (SNR) levels (-40dB to +20dB). The traditional electronic support measure (ESM) systems require high SNR for radar signal classification. LPI (low probability of intercept) radar signals that are received by ESM systems are usually corrupted by noise. So, we demonstrate through extensive simulations that it is possible to achieve high classification performance at low SNR levels providing that the underlying radar signals are known in advance. MF bank classifier, 1D Convolutional Neural Networks (CNNs) and the minimum distance classifier using spectral-domain features (the skewness, the kurtosis, and the energy of the dominant frequency) have been derived for the radar signal classification and their performances have been compared with each other and with the optimal classifier. Scopus

The performance of software defect prediction(SDP) models is known to be dependent on the datasets used for training the models. Evolving data in a dynamic software development environment such as significant refactoring and organizational changes introduces new concept to the prediction model, thus making improved classification performance difficult. In this study, we investigate and assess the existence and impact of concept drift on SDP performances. We empirically asses the prediction performance of five models by conducting cross-version experiments using fifty-five releases of five open-source projects. Prediction performance fluctuated as the training datasets changed over time. Our results indicate that the quality and the reliability of defect prediction models fluctuate over time and that this instability should be considered by software quality teams when using historical datasets. The performance of a static predictor constructed with data from historical versions may degrade over time due to the challenges posed by concept drift. © 2020 IEEE. open access

In this work, we propose to use learned features for terrain classification of Polarimetric Synthetic Aperture Radar (PolSAR) images. In the proposed classification framework, the learned features are extracted from sliding window regions using Convolutional Neural Networks (CNNs), and then they are used for the classification with the linear Support Vector Machine (SVM) classifier. The classification performance of the proposed approach is compared with numerous target decomposition theorems (TDs) as the engineered features tested with two classifiers: Collective Network of Binary Classifiers (CNBCs) and SVMs. The experimental evaluations over two commonly used benchmark AIRSAR PolSAR images, San Francisco Bay and Flevoland at L-Band, reveal that the classification performance of the learned features with CNNs outperforms the performance of the engineered features as TDs even the dimension of learned features is the quarter of the engineered features. acceptedVersion

Diverse Discrete Discriminant Analysis (DDA) models perform differently in different samples. This fact has encouraged research in combined models which seems particularly promising when the a priori classes are not well separated or when small or moderate sized samples are considered, which often occurs in practice. In this study, we evaluate the performance of a convex combination of two DDA models: the First-Order Independence Model (FOIM) and the Dependence Trees Model (DTM). We use simulated data sets with two classes and consider diverse data complexity factors which may influence performance of the combined model -the separation of classes, balance, and number of missing states, as well as sample size and also the number of parameters to be estimated in DDA. We resort to cross-validation to evaluate the precision of classification. The results obtained illustrate the advantage of the proposed combination when compared with FOIM and DTM: it yields the best results, especially when very small samples are considered. The experimental study also provides a ranking of the data complexity factors, according to their relative impact on classification performance, by means of a regression model. It leads to the conclusion that the separation of classes is the most influential factor in classification performance. The ratio between the number of degrees of freedom and sample size, along with the proportion of missing states in the minority class, also has significant impact on classification performance. An additional gain of this study, also deriving from the estimated regression model, is the ability to successfully predict the precision of classification in a real data set based on the data complexity factors. info:eu-repo/semantics/publishedVersion

WOS: 000386750300004, 28110727, Congestive heart failure (CHF) is a degree of cardiac disease occurring as a result of the heart's inability to pump enough blood for the human body. In recent studies, coronary artery disease (CAD) is accepted as the most important cause of CHF. This study focuses on the diagnosis of both the CHF and the CAD. The Hilbert-Huang transform (HHT), which is effective on nonlinear and non-stationary signals, is used to extract the features from R-R intervals obtained from the raw electrocardiogram data. The statistical features are extracted from instinct mode functions that are obtained applying the HHT to R-R intervals. Classification performance is examined with extracted statistical features using a multilayer perceptron neural network. The designed model classified the CHF, the CAD patients and a normal control group with rates of 97.83%, 93.79% and 100%, accuracy, specificity and sensitivity, respectively. Also, early diagnosis of the CHF was performed by interpretation of the CAD with a classification accuracy rate of 97.53%, specificity of 98.18% and sensitivity of 97.13%. As a result, a single system having the ability of both diagnosis and early diagnosis of CHF is performed by integrating the CAD diagnosis method to the CHF diagnosis method. (C) 2016 Elsevier Ireland Ltd. All rights reserved.

Medical Technologies Congress (TIPTEKNO) -- OCT 03-05, 2019 -- Izmir, TURKEY, WOS: 000516830900058, Parkinson's disease can be diagnosed by the speech signals. In general, the data obtained by feature extraction algorithms from the speech signals are used in any classification algorithm. Some of the extracted features have a high ability to represent the relevant problem, while others are low. In the diagnosis of Parkinson's disease, it is very important to determine which of the extracted features from the speech signals may increase the classification performance. In this paper, Artificial Bee Colony algorithm based feature selection approach is proposed for the solution of the mentioned problem. The proposed method has been analyzed in comparison with the well-known classification methods including support vector machine, k nearest neighbor, Naive Bayesian, decision tree., Biyomedikal Klinik Muhendisligi Dernegi, Izmir Katip Celebi Univ, Biyomedikal Muhendisligi Bolumu

WOS: 000471669900024, 31125914, Convolutional neural networks with strong representation ability of deep structures have ever increasing popularity in many research areas. The main difference of Convolutional Neural Networks with respect to existing similar artificial neural networks is the inclusion of the convolutional part. This inclusion directly increases the performance of artificial neural networks. This fact has led to the development of many different convolutional models and techniques. In this work, a novel convolution technique named as Differential Convolution and updated error back-propagation algorithm is proposed. The proposed technique aims to transfer feature maps containing directional activation differences to the next layer. This implementation takes the idea of how convolved features change on the feature map into consideration. In a sense, this process adapts the mathematical differentiation operation into the convolutional process. Proposed improved back propagation algorithm also considers neighborhood activation errors. This property increases the classification performance without changing the number of filters. Four different experiment sets were performed to observe the performance and the adaptability of the differential convolution technique. In the first experiment set utilization of the differential convolution on a traditional convolutional neural network structure made a performance boost up to 55.29% for the test accuracy. In the second experiment set differential convolution adaptation raised the top1 and top5 test accuracies of AlexNet by 5.3% and 4.75% on ImageNet dataset. In the third experiment set differential convolution utilized model outperformed all compared convolutional structures. In the fourth experiment set, the Differential VGGNet model obtained by adapting proposed differential convolution technique performed 93.58% and 75.06% accuracy values for CIFAR10 and CI-FAR100 datasets, respectively. The accuracy values of the Differential NIN model containing differential convolution operation were 92.44% and 72.65% for the same datasets. In these experiment sets, it was observed that the differential convolution technique outperformed both traditional convolution and other compared convolution techniques. In addition, easy adaptation of the proposed technique to different convolutional structures and its efficiency demonstrate that popular deep learning models may be improved with differential convolution. (C) 2019 Elsevier Ltd. All rights reserved.

The ALFA project aims at timely detection, tracking, classification, and intent assessment of LSS targets. The system relies on a heterogeneous sensor suite, including radar. The objective of the radar component is sector surveillance including target classification. Since the revisit time needs to be short, classification must be done with very short time-on-target. Based on measurements, three suitable features for classification of two relevant target classes, i.e., small aircraft and helicopters, have been developed. These features exploit the targets' micro-Doppler characteristics and their evolution over time. Best classification performance is obtained by using a combination of these features and by considering the variation of the features' distributions depending on the signal-to-noise ratio. © 2019 EuMA.

Deep learning with convolutional neural networks (ConvNets) have dramatically improved learning capabilities of computer vision applications just through considering raw data without any prior feature extraction. Nowadays, there is rising curiosity in interpreting and analyzing electroencephalography (EEG) dynamics with ConvNets. Our study focused on ConvNets of different structures, constructed for predicting imagined left and right movements on a subject-independent basis through raw EEG data. Results showed that recently advanced methods in machine learning field, i.e. adaptive moments and batch normalization together with dropout strategy, improved ConvNets predicting ability, outperforming that of conventional fully-connected neural networks with widely-used spectral features., 5 pages, 2 figures, springer

WOS: 000457264700008, 30712606, Background and objective: ECG is one of the biometric signals that has been studied in peer-reviewed over past years. The developments on the signal analysis methods show that the studies on the ECG would continue unabatedly. It has a common use on cardiac diseases with high rates of classification performances by integrating it with signal analysis methods. The aim of the study is to utilize the ECG for human identification. Methods: Second Order Difference Plot (SODP) is a non-linear time-series analysis method that allows determining the features using the statistical analysis of the wave distributions. The SODP features were extracted using different quantification methods for ECG-based human identification. A new quantification approach has been proposed on the SODP for ECG-based human identification. The proposed method, Logarithmic Grid Analysis, was compared with the existing quantification methods on the SODP. The region of the SODP was divided into sub-regions with logarithmically increasing distances and the numbers of data points in each logarithmic sub regions were calculated in the proposed method. Three different databases were used to test the validity of the method. These records have been tested with the conventional feature extraction methods on the SODP. The long-term ECG signals were divided into 5-s short-term ECG signals. Results: The Logarithmic Grid Analysis features that were counted from short-time ECG signals were classified with k-Nearest Neighbor algorithm using 10-fold cross validation, and the identification performance of the proposed model was evaluated. Consequently, high accuracy rates of 91.96%, 99.86% and 95.12% were achieved on ECG-based human identification using the Logarithmic Grid Analysis method on the SODP. Conclusions: The density score of data points at the center of the SODP is too high. This case increases the importance of the regions close the center in order to find the detailed and significant features from the SODP. The number of data points at the center has been extracted in more detail and the vertex areas of the major axes of the SODP can be interpreted in the aggregate sub-regions by using logarithmically increasing distances with a small number of feature size. (C) 2019 Elsevier B.V. All rights reserved.

Multivariate Classification of Brugada Syndrome Patients Based on the Autonomic Response During Sleep, Exercise and Head-up Tilt Testing Several autonomic markers were estimated overnight and during exercise and head-up tilt (HUT) testing for 44 BS patients, to design classifiers capable of distinguishing patients at different levels of risk. The classification performance of predictive models built from the optimization of a step-based machine-learning method were compared, so as to identify those autonomic protocols and markers best distinguishing between symptomatic and asymptomatic patients. Although exercise and HUT testing together led to better predictive results than when they were separately assessed, among all analyzed combinations, the night-based classifier presented the best performance (AUC = 95%), using the least amount of features. This optimal features subset was mostly composed of markers extracted between 4 a.m. - 5 a.m. Thus, results provide further evidence for the role of nighttime analysis, mainly during the last hours of sleep, for risk stratification in BS

Feature selection is an important issue when one deals with large amounts of omics data. Feature selection offers interpretability of the predictive model to the domain expert. Such interpretability is strongly affected by the typical instability of current feature selection methods. Instability here refers to the fact that the selected features may be drastically different even after marginal modification of the data. In this paper, we investigate the possibility of a tradeoff between the classification performance and the stability of a standard feature selection method: the Recursive Feature Elimination algorithm (RFE). The compromise is done by explicitly favoring the selection of some features through differential shrinkage. Such an approach allows the domain expert to control explicitly the trade-off between selection stability and predictive accuracy. Domain experts can thus select particular Pareto-optimal compromises, based on their personal preferences. As a secondary contribution, we propose the use of the hypervolume metric to assess the performance of methods realizing such a compromise and we define a corresponding confidence interval. Our approach is evaluated on prostate cancer diagnosis from microarray data and handwritten digit recognition tasks. Results show that the aforementioned tradeoff is effectively possible and that prior knowledge is an efficient way of stabilizing the selection.

The ALFA project aims at timely detection, tracking, classification, and intent assessment of LSS targets. The system relies on a heterogeneous sensor suite, including radar. The objective of the radar component is sector surveillance including target classification. Since the revisit time needs to be short, classification must be done with very short time-on-target. Based on measurements, three suitable features for classification of two relevant target classes, i.e., small aircraft and helicopters, have been developed. These features exploit the targets' micro-Doppler characteristics and their evolution over time. Best classification performance is obtained by using a combination of these features and by considering the variation of the features' distributions depending on the signal-to-noise ratio. © 2019 EuMA.

Cyclone separators are widely used for dedusting of particle-laden flows in many branches of industry. A new cyclone containing several collectors was designed in modular structure to investigate the capability of particle classification of the cyclone during particle separation process. The particulate air was supplied into the cyclone in a test bench operated in suction mode under ambient temperature and atmospheric pressure conditions. The flow rate and pressure drop were measured, and the particles accumulated in the dust collectors were weighed and analyzed for each test. The changes in the overall collection efficiency and pressure drop with the flow rate were obtained. In order to determine the classification characteristics of the cyclone, the experiments were repeated in different configurations obtained by changing the locations and number of the collectors. The mean particle diameter and particle size distribution of the dust collected in each collector were acquired. Analysis of the results shows that the designed cyclone is capable of classification incoming particles into 2, 3 or 4 classes in the cyclone according to particle size. Classification performance of the configurations and influences of flow rate on this performance were also investigated. A distinct and satisfactory classification was obtained in the configuration with three collectors.

An approach exploiting the principles of Receiver Operating Characteristic (ROC) curves for the simultaneous optimization of both the complexity and the decision threshold in Soft Independent Modeling of Class Analogy (SIMCA) classification models is here proposed. The outcomes resulting from the analysis of two simulated and four real case-studies highlight that, in the presence of strong overlap among various categories of samples, the implemented method can lead to better classification efficiency in external validation, compared to fixing such a threshold a priori. This guarantees a higher robustness toward class dispersion. On the other hand, in cases of clearer and more definite separation among the different groups of observations, their classification performance is equally satisfactory for test samples. ispartof: ANALYTICAL CHEMISTRY vol:90 issue:18 pages:10738-10747 ispartof: location:United States status: published

Derin evrişimsel sinir ağları (ESA) pek çok sınıflandırma probleminde olduğu gibi anlambilimsel görüntü bölütlemede de çok ciddi başarı göstermiştir. Fakat derin ağların eğitilmesi hem zaman alıcıdır hem de geniş bir eğitim veri kümesine ihtiyaç duymaktadır. Bir veri kümesinde eğitilen ağın başka bir görev ya da veri kümesine uygulanabilmesi için transfer öğrenme ile yeniden eğitilmesi gerekmektedir. Transfer öğrenmeye alternatif olarak ağ katmanlarından çıkarılan öznitelik vektörleri doğrudan sınıflandırma amaçlı kullanılabilir. Bu bildiride genel ESA mimarilerinden elde edilen özniteliklerin eğitim gerektirmeyen bir görüntü etiketleme yönteminde kullanılmasının sınıflandırma başarımına katkısı incelenmiştir. Derin ağlarda ‘öğrenilmiş’ öznitelikler ile SIFT gibi ‘el yapımı’ özniteliklerin birlikte kullanılmasının etiketleme doğruluğunu artırdığı gösterilmiştir. Varolan ön eğitimli ağların kullanılması sayesinde önerilen yaklaşım herhangi bir veri kümesinde yeniden eğitime gerek olmadan kolayca uygulanabilmektedir. Önerilen yöntem iki veri kümesinde test edilmiş ve etiketleme doğruluğu benzer yöntemlerle karşılaştırmalı olarak sunulmuştur. Deep convolutional neural networks (CNN) have shown significant success in many classification problems including semantic image segmentation. However training of deep networks is time consuming and requires large training datasets. A network trained in one dataset could be applied to another task or dataset through transfer learning and retraining. As an alternative to transfer learning, feature vectors that are extracted from network layers could be directly used for classification purposes. In this paper we investigate the improvement in classification performance when features extracted from generic CNN architectures are used in an image labeling algorithm that does not require training. We show that the use of 'learned' features from deep networks together with 'hand-crafted' features such as SIFT increases the labeling accuracy. Since existing pre-trained networks are used, the proposed approach could be easily applied to any dataset without any retraining. The proposed method is tested in two datasets and labeling accuracies are compared with similar existing methods. Publisher's Version

In this paper a suitable methodology for the improvement of the reliability of results in classification systems based on 3D images is proposed. More in detail, it is based on the knowledge of the uncertainty of the features constituting the 3D image (obtained processing a pair of two 2D stereoscopic images) and on a suitable statistical approach providing a confidence level to the classification result. These pieces of information are then managed in order to improve the classification performance in terms of correct classification and false reject percentages. The experimental results, obtained applying the methodology on an Active Appearance Models algorithm for feature detection and triangulating the 3D features, show that, compared with a basic approach (which generally does not take into account the uncertainty on 3D features), the proposed methodology allows to significantly improve the classification performance even in scenarios characterized by a high uncertainty.