Abstract: In order to be able to automatically differentiate
between two modes of permanent flow of a liquid simulating blood,
it was imperative to put together a data bank. Thus, the acquisition of
the various amplitude spectra of the Doppler signal of this liquid in
laminar flow and other spectra in turbulent flow enabled us to
establish an automatic difference between the two modes. According
to the number of parameters and their nature, a comparative study
allowed us to choose the best classifier.
Abstract: In this paper we propose a new approach for flexible document categorization according to the document type or genre instead of topic. Our approach implements two homogenous classifiers: contextual classifier and logical classifier. The contextual classifier is based on the document URL, whereas, the logical classifier use the logical structure of the document to perform the categorization. The final categorization is obtained by combining contextual and logical categorizations. In our approach, each document is assigned to all predefined categories with different membership degrees. Our experiments demonstrate that our approach is best than other genre categorization approaches.
Abstract: In this work, we present a novel active learning approach
for learning a visual object detection system. Our system
is composed of an active learning mechanism as wrapper around
a sub-algorithm which implement an online boosting-based learning
object detector. In the core is a combination of a bootstrap procedure
and a semi automatic learning process based on the online boosting
procedure. The idea is to exploit the availability of classifier during
learning to automatically label training samples and increasingly
improves the classifier. This addresses the issue of reducing labeling
effort meanwhile obtain better performance. In addition, we propose
a verification process for further improvement of the classifier.
The idea is to allow re-update on seen data during learning for
stabilizing the detector. The main contribution of this empirical study
is a demonstration that active learning based on an online boosting
approach trained in this manner can achieve results comparable or
even outperform a framework trained in conventional manner using
much more labeling effort. Empirical experiments on challenging data
set for specific object deteciton problems show the effectiveness of
our approach.
Abstract: Effectiveness of Artificial Neural Networks (ANN)
and Support Vector Machines (SVM) classifiers for fault diagnosis of
rolling element bearings are presented in this paper. The
characteristic features of vibration signals of rotating driveline that
was run in its normal condition and with faults introduced were used
as input to ANN and SVM classifiers. Simple statistical features such
as standard deviation, skewness, kurtosis etc. of the time-domain
vibration signal segments along with peaks of the signal and peak of
power spectral density (PSD) are used as features to input the ANN
and SVM classifier. The effect of preprocessing of the vibration
signal by Discreet Wavelet Transform (DWT) prior to feature
extraction is also studied. It is shown from the experimental results
that the performance of SVM classifier in identification of bearing
condition is better then ANN and pre-processing of vibration signal
by DWT enhances the effectiveness of both ANN and SVM classifier
Abstract: Sleep stage scoring is the process of classifying the
stage of the sleep in which the subject is in. Sleep is classified into
two states based on the constellation of physiological parameters.
The two states are the non-rapid eye movement (NREM) and the
rapid eye movement (REM). The NREM sleep is also classified into
four stages (1-4). These states and the state wakefulness are
distinguished from each other based on the brain activity. In this
work, a classification method for automated sleep stage scoring
based on a single EEG recording using wavelet packet decomposition
was implemented. Thirty two ploysomnographic recording from the
MIT-BIH database were used for training and validation of the
proposed method. A single EEG recording was extracted and
smoothed using Savitzky-Golay filter. Wavelet packets
decomposition up to the fourth level based on 20th order Daubechies
filter was used to extract features from the EEG signal. A features
vector of 54 features was formed. It was reduced to a size of 25 using
the gain ratio method and fed into a classifier of regression trees. The
regression trees were trained using 67% of the records available. The
records for training were selected based on cross validation of the
records. The remaining of the records was used for testing the
classifier. The overall correct rate of the proposed method was found
to be around 75%, which is acceptable compared to the techniques in
the literature.
Abstract: This paper reports a new approach on identifying the
individuality of persons by using parametric classification of multiple
mental thoughts. In the approach, electroencephalogram (EEG)
signals were recorded when the subjects were thinking of one or
more (up to five) mental thoughts. Autoregressive features were
computed from these EEG signals and classified by Linear
Discriminant classifier. The results here indicate that near perfect
identification of 400 test EEG patterns from four subjects was
possible, thereby opening up a new avenue in biometrics.
Abstract: Text data mining is a process of exploratory data
analysis. Classification maps data into predefined groups or classes.
It is often referred to as supervised learning because the classes are
determined before examining the data. This paper describes proposed
radial basis function Classifier that performs comparative crossvalidation
for existing radial basis function Classifier. The feasibility
and the benefits of the proposed approach are demonstrated by means
of data mining problem: direct Marketing. Direct marketing has
become an important application field of data mining. Comparative
Cross-validation involves estimation of accuracy by either stratified
k-fold cross-validation or equivalent repeated random subsampling.
While the proposed method may have high bias; its performance
(accuracy estimation in our case) may be poor due to high variance.
Thus the accuracy with proposed radial basis function Classifier was
less than with the existing radial basis function Classifier. However
there is smaller the improvement in runtime and larger improvement
in precision and recall. In the proposed method Classification
accuracy and prediction accuracy are determined where the
prediction accuracy is comparatively high.
Abstract: This paper proposes an efficient method to classify
inverse synthetic aperture (ISAR) images. Because ISAR images can
be translated and rotated in the 2-dimensional image place, invariance
to the two factors is indispensable for successful classification. The
proposed method achieves invariance to translation and rotation of
ISAR images using a combination of two-dimensional Fourier
transform, polar mapping and correlation-based alignment of the
image. Classification is conducted using a simple matching score
classifier. In simulations using the real ISAR images of five scaled
models measured in a compact range, the proposed method yields
classification ratios higher than 97 %.
Abstract: The study of the interaction between humans and
computers has been emerging during the last few years. This
interaction will be more powerful if computers are able to perceive
and respond to human nonverbal communication such as emotions. In
this study, we present the image-based approach to emotion
classification through lower facial expression. We employ a set of
feature points in the lower face image according to the particular face
model used and consider their motion across each emotive expression
of images. The vector of displacements of all feature points input to
the Adaptive Support Vector Machines (A-SVMs) classifier that
classify it into seven basic emotions scheme, namely neutral, angry,
disgust, fear, happy, sad and surprise. The system was tested on the
Japanese Female Facial Expression (JAFFE) dataset of frontal view
facial expressions [7]. Our experiments on emotion classification
through lower facial expressions demonstrate the robustness of
Adaptive SVM classifier and verify the high efficiency of our
approach.
Abstract: Although face recognition seems as an easy task for
human, automatic face recognition is a much more challenging task
due to variations in time, illumination and pose. In this paper, the
influence of time-lapse on visible and thermal images is examined.
Orthogonal moment invariants are used as a feature extractor to
analyze the effect of time-lapse on thermal and visible images and the
results are compared with conventional Principal Component
Analysis (PCA). A new triangle square ratio criterion is employed
instead of Euclidean distance to enhance the performance of nearest
neighbor classifier. The results of this study indicate that the ideal
feature vectors can be represented with high discrimination power
due to the global characteristic of orthogonal moment invariants.
Moreover, the effect of time-lapse has been decreasing and enhancing
the accuracy of face recognition considerably in comparison with
PCA. Furthermore, our experimental results based on moment
invariant and triangle square ratio criterion show that the proposed
approach achieves on average 13.6% higher in recognition rate than
PCA.
Abstract: This work presents an approach for the construction of a hybrid color-texture space by using mutual information. Feature extraction is done by the Laws filter with SVM (Support Vectors Machine) as a classifier. The classification is applied on the VisTex database and a SPOT HRV (XS) image representing two forest areas in the region of Rabat in Morocco. The result of classification obtained in the hybrid space is compared with the one obtained in the RGB color space.
Abstract: This paper explores the effectiveness of machine
learning techniques in detecting firms that issue fraudulent financial
statements (FFS) and deals with the identification of factors
associated to FFS. To this end, a number of experiments have been
conducted using representative learning algorithms, which were
trained using a data set of 164 fraud and non-fraud Greek firms in the
recent period 2001-2002. The decision of which particular method to
choose is a complicated problem. A good alternative to choosing
only one method is to create a hybrid forecasting system
incorporating a number of possible solution methods as components
(an ensemble of classifiers). For this purpose, we have implemented
a hybrid decision support system that combines the representative
algorithms using a stacking variant methodology and achieves better
performance than any examined simple and ensemble method. To
sum up, this study indicates that the investigation of financial
information can be used in the identification of FFS and underline the
importance of financial ratios.
Abstract: In this paper we present a new approach to detecting a
flaw in T.O.F.D (Time Of Flight Diffraction) type ultrasonic image
based on texture features. Texture is one of the most important
features used in recognizing patterns in an image. The paper
describes texture features based on 2D Gabor functions, i.e.,
Gaussian shaped band-pass filters, with dyadic treatment of the radial
spatial frequency range and multiple orientations, which represent an
appropriate choice for tasks requiring simultaneous measurement in
both space and frequency domains. The most relevant features are
used as input data on a Fuzzy c-mean clustering classifier. The
classes that exist are only two: 'defects' or 'no defects'. The proposed
approach is tested on the T.O.F.D image achieved at the laboratory
and on the industrial field.
Abstract: The problem of spam has been seriously troubling the Internet community during the last few years and currently reached an alarming scale. Observations made at CERN (European Organization for Nuclear Research located in Geneva, Switzerland) show that spam mails can constitute up to 75% of daily SMTP traffic. A naïve Bayesian classifier based on a Bag Of Words representation of an email is widely used to stop this unwanted flood as it combines good performance with simplicity of the training and classification processes. However, facing the constantly changing patterns of spam, it is necessary to assure online adaptability of the classifier. This work proposes combining such a classifier with another NBC (naïve Bayesian classifier) based on pairs of adjacent words. Only the latter will be retrained with examples of spam reported by users. Tests are performed on considerable sets of mails both from public spam archives and CERN mailboxes. They suggest that this architecture can increase spam recall without affecting the classifier precision as it happens when only the NBC based on single words is retrained.
Abstract: Genetic Folding (GF) a new class of EA named as is
introduced for the first time. It is based on chromosomes composed
of floating genes structurally organized in a parent form and
separated by dots. Although, the genotype/phenotype system of GF
generates a kernel expression, which is the objective function of
superior classifier. In this work the question of the satisfying
mapping-s rules in evolving populations is addressed by analyzing
populations undergoing either Mercer-s or none Mercer-s rule. The
results presented here show that populations undergoing Mercer-s
rules improve practically models selection of Support Vector
Machine (SVM). The experiment is trained multi-classification
problem and tested on nonlinear Ionosphere dataset. The target of this
paper is to answer the question of evolving Mercer-s rule in SVM
addressed using either genetic folding satisfied kernel-s rules or not
applied to complicated domains and problems.
Abstract: Text categorization is the problem of classifying text
documents into a set of predefined classes. In this paper, we
investigated three approaches to build a meta-classifier in order to
increase the classification accuracy. The basic idea is to learn a metaclassifier
to optimally select the best component classifier for each
data point. The experimental results show that combining classifiers
can significantly improve the accuracy of classification and that our
meta-classification strategy gives better results than each individual
classifier. For 7083 Reuters text documents we obtained a
classification accuracies up to 92.04%.
Abstract: In this paper we designed and implemented a new
ensemble of classifiers based on a sequence of classifiers which were
specialized in regions of the training dataset where errors of its
trained homologous are concentrated. In order to separate this
regions, and to determine the aptitude of each classifier to properly
respond to a new case, it was used another set of classifiers built
hierarchically. We explored a selection based variant to combine the
base classifiers. We validated this model with different base
classifiers using 37 training datasets. It was carried out a statistical
comparison of these models with the well known Bagging and
Boosting, obtaining significantly superior results with the
hierarchical ensemble using Multilayer Perceptron as base classifier.
Therefore, we demonstrated the efficacy of the proposed ensemble,
as well as its applicability to general problems.
Abstract: In this paper we address the problem of musical style
classification, which has a number of applications like indexing in
musical databases or automatic composition systems. Starting from
MIDI files of real-world improvisations, we extract the melody track
and cut it into overlapping segments of equal length. From these
fragments, some numerical features are extracted as descriptors of
style samples. We show that a standard Bayesian classifier can be
conveniently employed to build an effective musical style classifier,
once this set of features has been extracted from musical data.
Preliminary experimental results show the effectiveness of the
developed classifier that represents the first component of a musical
audio retrieval system
Abstract: The Linear discriminant analysis (LDA) can be
generalized into a nonlinear form - kernel LDA (KLDA) expediently
by using the kernel functions. But KLDA is often referred to a general
eigenvalue problem in singular case. To avoid this complication, this
paper proposes an iterative algorithm for the two-class KLDA. The
proposed KLDA is used as a nonlinear discriminant classifier, and the
experiments show that it has a comparable performance with SVM.
Abstract: In this paper, we present a new method for
incorporating global shift invariance in support vector machines.
Unlike other approaches which incorporate a feature extraction stage,
we first scale the image and then classify it by using the modified
support vector machines classifier. Shift invariance is achieved by
replacing dot products between patterns used by the SVM classifier
with the maximum cross-correlation value between them. Unlike the
normal approach, in which the patterns are treated as vectors, in our
approach the patterns are treated as matrices (or images). Crosscorrelation
is computed by using computationally efficient
techniques such as the fast Fourier transform. The method has been
tested on the ORL face database. The tests indicate that this method
can improve the recognition rate of an SVM classifier.