Wyniki wyszukiwania - Biblioteka Nauki

1

Correlation-based feature selection strategy in classification problems

100%

Michalak K. , Kwaśnicka H.

|

nr 4

503-511

EN

In classification problems, the issue of high dimensionality, of data is often considered important. To lower data dimensionality, feature selection methods are often employed. To select a set of features that will span a representation space that is as good as possible for the classification task, one must take into consideration possible interdependencies between the features. As a trade-off between the complexity of the selection process and the quality of the selected feature set, a pairwise selection strategy has been recently suggested. In this paper, a modified pairwise selection strategy is proposed. Our research suggests that computation time can be significantly lowered while maintaining the quality of the selected feature sets by using mixed univariate and bivariate feature evaluation based on the correlation between the features. This paper presents the comparison of the performance of our method with that of the unmodified pairwise selection strategy based on several well-known benchmark sets. Experimental results show that, in most cases, it is possible to lower computation time and that with high statistical significance the quality of the selected feature sets is not lower compared with those selected using the unmodified pairwise selection process.

2

Supervised Kernel Principal Component Analysis by Most Expressive Feature Reordering

100%

Ślot K. , Adamiak K. , Duch P. , Żurek D.

Journal of Telecommunications and Information Technology

|

2015

|

tom nr 2

3--10

EN

The presented paper is concerned with feature space derivation through feature selection. The selection is performed on results of kernel Principal Component Analysis (kPCA) of input data samples. Several criteria that drive feature selection process are introduced and their performance is assessed and compared against the reference approach, which is a combination of kPCA and most expressive feature reordering based on the Fisher linear discriminant criterion. It has been shown that some of the proposed modifications result in generating feature spaces with noticeably better (at the level of approximately 4%) class discrimination properties.

3

Geometryczna metoda selekcji informacji diagnostycznej

100%

Dybała J. , Radkowski S.

Diagnostyka

|

2004

|

tom Vol. 30, T. 1

159--162

PL

W pracy przedstawiono metodę selekcji cech stanu obiektu opartą na geometrii przestrzeni obserwacji. Zaprezentowana metoda selekcji informacji wykorzystuje dwa kryteria: zmodyfiko-wane kryterium Sebestyena oraz oryginalne kryterium liczby wzorców klas.

EN

In the paper the feature selection method of object state is presented. This method is based on geometry of the observation space. The presented method of the information selection uses two criteria: the modified Sebestyen's criterion and original criterion of the prototypes classes number.

4

Pattern classification by spiking neural networks combining self-organized and reward-related spike-timing-dependent plasticity

100%

Nobukawa S. , Nishimura H. , Yamanishi T.

|

2019

|

tom Vol. 9, No. 4

283--291

EN

Many recent studies have applied to spike neural networks with spike-timing-dependent plasticity (STDP) to machine learning problems. The learning abilities of dopaminemodulated STDP (DA-STDP) for reward-related synaptic plasticity have also been gathering attention. Following these studies, we hypothesize that a network structure combining self-organized STDP and reward-related DA-STDP can solve the machine learning problem of pattern classification. Therefore, we studied the ability of a network in which recurrent spiking neural networks are combined with STDP for non-supervised learning, with an output layer joined by DA-STDP for supervised learning, to perform pattern classification. We confirmed that this network could perform pattern classification using the STDP effect for emphasizing features of the input spike pattern and DA-STDP supervised learning. Therefore, our proposed spiking neural network may prove to be a useful approach for machine learning problems.

5

Data Stream Classification Using Classifier Ensemble

100%

Woźniak M. , Kasprzak A.

|

tom Vol. 23

21--32

EN

For the contemporary business, the crucial factor is making smart decisions on the basis of the knowledge hidden in stored data. Unfortunately, traditional simple methods of data analysis are not sufficient for efficient management of modern enterprizes, because they are not appropriate for the huge and growing amount of the data stored by them. Additionally data usually comes continuously in the form of so-called data stream. The great disadvantage of traditional classification methods is that they assume that statistical properties of the discovered concept are being unchanged, while in real situation, we could observe so-called concept drift, which could be caused by changes in the probabilities of classes or/and conditional probability distributions of classes. The potential for considering new training data is an important feature of machine learning methods used in security applications (spam filtering or intrusion detection) or decision support systems for marketing departments, which need to follow the changing client behavior. Unfortunately, the occurrence of concept drift dramatically decreases classification accuracy. This work presents the comprehensive study on the ensemble classifier approach applied to the problem of drifted data streams. Especially it reports the research on modifications of previously developed Weighted Aging Classifier Ensemble (WAE) algorithm, which is able to construct a valuable classifier ensemble for classification of incremental drifted stream data. We generalize WAE method and propose the general framework for this approach. Such framework can prune an classifier ensemble before or after assigning weights to individual classifiers. Additionally, we propose new classifier pruning criteria, weight calculation methods, and aging operators. We also propose rejuvenating operator, which is able to soften the aging effect, which could be useful, especially in the case if quite ”old” classifiers are high quality models, i.e., their presence increases ensemble accuracy, what could be found, e.g., in the case of recurring concept drift. The chosen characteristics of the proposed frameworks were evaluated on the basis of the wide range of computer experiments carried out on the two benchmark data streams. Obtained results confirmed the usability of proposed method to the data stream classification with the presence of incremental concept drift.

6

A Note on a priori Estimations of Classification Circuit Complexity

88%

Albrecht A. A. , Chashkin A. V. , Iliopoulos C. S. , Kasim-Zade O. M. , Lappas G. , Steinhöfel K. K.

|

2010

|

tom Vol. 104, nr 3

201-217

EN

The paper aims at tight upper bounds on the size of pattern classification circuits that can be used for a priori parameter settings in a machine learning context. The upper bounds relate the circuit size S(C) to n_L := .log_2mL., where mL is the number of training samples. In particular, we show that there exist unbounded fan-in threshold circuits with less than (a) [formula] gates for unbounded depth, (b) SL [formula] gates for small bounded depth, where in both cases all mL samples are classified correctly. We note that the upper bounds do not depend on the length n of input (sample) vectors. Since n_L << n in real-world problem settings, the upper bounds return values that are suitable for practical applications. We provide experimental evidence that the circuit size estimations work well on a number of pattern classification tasks. As a result, we formulate the conjecture that [formula] gates are sufficient to achieve a high generalization rate of bounded-depth classification circuits.

7

Critical exponent analysis applied to surface EMG signals for multifunction myoelectric control

75%

Phinyomark A. , Phothisonothai M. , Phukpattaranont P. , Limsakul C.

|

tom Vol. 18, nr 4

645-658

EN

Based on recent advances in non-linear analysis, the surface electromyography (sEMG) signal has been studied from the viewpoints of self-affinity and complexity. In this study, we examine usage of critical exponent analysis (CE) method, a fractal dimension (FD) estimator, to study properties of the sEMG signal and to deploy these properties to characterize different movements for gesture recognition. SEMG signals were recorded from thirty subjects with seven hand movements and eight muscle channels. Mean values and coefficient of variations of the CE from all experiments show that there are larger variations between hand movement types but there is small variation within the same type. It also shows that the CE feature related to the self-affine property for the sEMG signal extracted from different activities is in the range of 1.855∼2.754. These results have also been evaluated by analysis-of-variance (p-value). Results show that the CE feature is more suitable to use as a learning parameter for a classifier compared with other representative features including root mean square, median frequency and Higuchi's method. Most p-values of the CE feature were less than 0.0001. Thus the FD that is computed by the CE method can be applied to be used as a feature for a wide variety of sEMG applications.

8

Correlation-based feature selection strategy in classification problems

75%

Michalak K. , Kwaśnicka H.

|

2006

|

tom Vol. 16, no 4

503-511

EN

In classification problems, the issue of high dimensionality, of data is often considered important. To lower data dimensionality, feature selection methods are often employed. To select a set of features that will span a representation space that is as good as possible for the classification task, one must take into consideration possible interdependencies between the features. As a trade-off between the complexity of the selection process and the quality of the selected feature set, a pairwise selection strategy has been recently suggested. In this paper, a modified pairwise selection strategy is proposed. Our research suggests that computation time can be significantly lowered while maintaining the quality of the selected feature sets by using mixed univariate and bivariate feature evaluation based on the correlation between the features. This paper presents the comparison of the performance of our method with that of the unmodified pairwise selection strategy based on several well-known benchmark sets. Experimental results show that, in most cases, it is possible to lower computation time and that with high statistical significance the quality of the selected feature sets is not lower compared with those selected using the unmodified pairwise selection process.

9

A new method for system modelling and pattern classification

75%

Rutkowski L.

|

nr 1

10

Application of Two Dimensional Wavelet Transform for Classification of Power Quality Disturbances

75%

Mollayi N. , Mokhtari H.

|

tom Vol. 17, iss. 2

1--7

EN

Identification of voltage and current disturbances is an important task in power system monitoring and protection. In this paper, the application of two-dimensional wavelet transform for characterization of a wide variety range of power quality disturbances is discussed, and a new algorithm, based on image processing techniques is proposed for this purpose. A matrix is formed based on a specified number of cycles in such a way that the samples of voltage signal in each cycle form one row of that matrix. This matrix can be regarded as a two dimensional image. A two-dimensional wavelet transform is used to decompose the image into approximation and details, which contain low frequency and high frequency components along the rows and columns, respectively. Different disturbances result into different special patterns in detail images. By processing the detail images, specific features are defined which can suitably discriminate various types of disturbances. Combination of the feature generation algorithm and a classifier system leads to a smart system for classification of wide variety range of disturbances.

11

New optical character recognition method based on Hu invariant moments and weighted voting

63%

Kmieć M.

|

2011

|

tom Vol. 19, nr 1

33-50

EN

Optical character recognition is an important image processing task. Its aim is to enable computers to recognise graphic characters without human supervision. The process of optical symbol recognition is divided into two stages. First, certain features of the character undergoing recognition are extracted, and second, a match to them is searched for in the library of models. This paper looks at Hu invariant moments, a well established set of image features, and discusses their performance in optical character recognition. One approach to using Hu invariant moments in pattern recognition is using a metric function to find the pattern in the library of models, that is of the same class as the pattern considered. In this paper a new classification method is proposed that performs better than the classic method of metric function.

12

Analysis of extracted cardiotocographic signal features to improve automated prediction of fetal outcome

51%

Jeżewski M. , Czabański R. , Wróbel J. , Horoba K.

|

tom Vol. 30, no. 4

29-47

EN

Cardiotocographic monitoring based on automated analysis of the fetal heart rate (FHR) signal is widely used for fetal assessment. However, the conclusion generation system is still needed to improve the abnormal fetal outcome prediction. Classification of the signals according to the predicted fetal outcome by means of neural networks is presented in this paper. Multi-layer perceptron neural networks were learned through seventeen time-domain signal features extracted during computerized analysis of 749 traces from 103 patients. The analysis included estimation of the FHR baseline, detection of acceleration and deceleration patterns as well as measurement of the instantaneous FHR variability. All the traces were retrospectively verified by the real fetal outcome defined by newborn delivery data. Influence of numerical and categorical representation of the input signal features, different data sets during learning, and gestational age as additional information, were investigated. We achieved the best sensitivity and specificity for the neural networks fed with numerical input variables together with additional information on the gestational age in the categorical form.

13

Machine-learning at the service of plastic surgery: a case study evaluating facial attractiveness and emotions using R language

38%

Štěpánek L. , Měšťák J. , Kasal P.

|

2019

|

tom Vol. 18

107--112

EN

Since the plastic surgery should consider that facial impression is always dependent on current facial emotion, it came to be verified how precise classification of facial images into sets of defined facial emotions is.