Abstract: As emails communications have no consistent
authentication procedure to ensure the authenticity, we present an
investigation analysis approach for detecting forged emails based on
Random Forests and Naïve Bays classifiers. Instead of investigating
the email headers, we use the body content to extract a unique writing
style for all the possible suspects. Our approach consists of four main
steps: (1) The cybercrime investigator extract different effective
features including structural, lexical, linguistic, and syntactic
evidence from previous emails for all the possible suspects, (2) The
extracted features vectors are normalized to increase the accuracy
rate. (3) The normalized features are then used to train the learning
engine, (4) upon receiving the anonymous email (M); we apply the
feature extraction process to produce a feature vector. Finally, using
the machine learning classifiers the email is assigned to one of the
suspects- whose writing style closely matches M. Experimental
results on real data sets show the improved performance of the
proposed method and the ability of identifying the authors with a
very limited number of features.
Abstract: In the past years a lot of effort has been made in the
field of face detection. The human face contains important features
that can be used by vision-based automated systems in order to
identify and recognize individuals. Face location, the primary step of
the vision-based automated systems, finds the face area in the input
image. An accurate location of the face is still a challenging task.
Viola-Jones framework has been widely used by researchers in order
to detect the location of faces and objects in a given image. Face
detection classifiers are shared by public communities, such as
OpenCV. An evaluation of these classifiers will help researchers to
choose the best classifier for their particular need. This work focuses
of the evaluation of face detection classifiers minding facial
landmarks.
Abstract: This paper describes a method to improve the robustness of a face recognition system based on the combination of two compensating classifiers. The face images are preprocessed by the appearance-based statistical approaches such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). LDA features of the face image are taken as the input of the Radial Basis Function Network (RBFN). The proposed approach has been tested on the ORL database. The experimental results show that the LDA+RBFN algorithm has achieved a recognition rate of 93.5%
Abstract: In the recent past Learning Classifier Systems have
been successfully used for data mining. Learning Classifier System
(LCS) is basically a machine learning technique which combines
evolutionary computing, reinforcement learning, supervised or
unsupervised learning and heuristics to produce adaptive systems. A
LCS learns by interacting with an environment from which it
receives feedback in the form of numerical reward. Learning is
achieved by trying to maximize the amount of reward received. All
LCSs models more or less, comprise four main components; a finite
population of condition–action rules, called classifiers; the
performance component, which governs the interaction with the
environment; the credit assignment component, which distributes the
reward received from the environment to the classifiers accountable
for the rewards obtained; the discovery component, which is
responsible for discovering better rules and improving existing ones
through a genetic algorithm. The concatenate of the production rules
in the LCS form the genotype, and therefore the GA should operate
on a population of classifier systems. This approach is known as the
'Pittsburgh' Classifier Systems. Other LCS that perform their GA at
the rule level within a population are known as 'Mitchigan' Classifier
Systems. The most predominant representation of the discovered
knowledge is the standard production rules (PRs) in the form of IF P
THEN D. The PRs, however, are unable to handle exceptions and do
not exhibit variable precision. The Censored Production Rules
(CPRs), an extension of PRs, were proposed by Michalski and
Winston that exhibit variable precision and supports an efficient
mechanism for handling exceptions. A CPR is an augmented
production rule of the form: IF P THEN D UNLESS C, where
Censor C is an exception to the rule. Such rules are employed in
situations, in which conditional statement IF P THEN D holds
frequently and the assertion C holds rarely. By using a rule of this
type we are free to ignore the exception conditions, when the
resources needed to establish its presence are tight or there is simply
no information available as to whether it holds or not. Thus, the IF P
THEN D part of CPR expresses important information, while the
UNLESS C part acts only as a switch and changes the polarity of D
to ~D. In this paper Pittsburgh style LCSs approach is used for
automated discovery of CPRs. An appropriate encoding scheme is
suggested to represent a chromosome consisting of fixed size set of
CPRs. Suitable genetic operators are designed for the set of CPRs
and individual CPRs and also appropriate fitness function is proposed
that incorporates basic constraints on CPR. Experimental results are
presented to demonstrate the performance of the proposed learning
classifier system.
Abstract: The purpose of this paper is to demonstrate the ability
of a genetic programming (GP) algorithm to evolve a team of data
classification models. The GP algorithm used in this work is
“multigene" in nature, i.e. there are multiple tree structures (genes)
that are used to represent team members. Each team member assigns
a data sample to one of a fixed set of output classes. A majority vote,
determined using the mode (highest occurrence) of classes predicted
by the individual genes, is used to determine the final class
prediction. The algorithm is tested on a binary classification problem.
For the case study investigated, compact classification models are
obtained with comparable accuracy to alternative approaches.
Abstract: Automatic reading of handwritten cheque is a computationally
complex process and it plays an important role in financial
risk management. Machine vision and learning provide a viable
solution to this problem. Research effort has mostly been focused
on recognizing diverse pitches of cheques and demand drafts with an
identical outline. However most of these methods employ templatematching
to localize the pitches and such schemes could potentially
fail when applied to different types of outline maintained by the
bank. In this paper, the so-called outline problem is resolved by
a cheque information tree (CIT), which generalizes the localizing
method to extract active-region-of-entities. In addition, the weight
based density plot (WBDP) is performed to isolate text entities and
read complete pitches. Recognition is based on texture features using
neural classifiers. Legal amount is subsequently recognized by both
texture and perceptual features. A post-processing phase is invoked
to detect the incorrect readings by Type-2 grammar using the Turing
machine. The performance of the proposed system was evaluated
using cheque and demand drafts of 22 different banks. The test data
consists of a collection of 1540 leafs obtained from 10 different
account holders from each bank. Results show that this approach
can easily be deployed without significant design amendments.
Abstract: For the characterization of the weld defect region in the radiographic image, looking for features which are invariant regarding the geometrical transformations (rotation, translation and scaling) proves to be necessary because the same defect can be seen from several angles according to the orientation and the distance from the welded framework to the radiation source. Thus, panoply of geometrical attributes satisfying the above conditions is proposed and which result from the calculation of the geometrical parameters (surface, perimeter, etc.) on the one hand and the calculation of the different order moments, on the other hand. Because the large range in values of the raw features and taking into account other considerations imposed by some classifiers, the scaling of these values to lie between 0 and 1 is indispensable. The principal component analysis technique is used in order to reduce the number of the attribute variables in the aim to give better performance to the further defect classification.
Abstract: The recognition of handwritten numeral is an
important area of research for its applications in post office, banks
and other organizations. This paper presents automatic recognition of
handwritten Kannada numerals based on structural features. Five
different types of features, namely, profile based 10-segment string,
water reservoir; vertical and horizontal strokes, end points and
average boundary length from the minimal bounding box are used in
the recognition of numeral. The effect of each feature and their
combination in the numeral classification is analyzed using nearest
neighbor classifiers. It is common to combine multiple categories of
features into a single feature vector for the classification. Instead,
separate classifiers can be used to classify based on each visual
feature individually and the final classification can be obtained based
on the combination of separate base classification results. One
popular approach is to combine the classifier results into a feature
vector and leaving the decision to next level classifier. This method
is extended to extract a better information, possibility distribution,
from the base classifiers in resolving the conflicts among the
classification results. Here, we use fuzzy k Nearest Neighbor (fuzzy
k-NN) as base classifier for individual feature sets, the results of
which together forms the feature vector for the final k Nearest
Neighbor (k-NN) classifier. Testing is done, using different features,
individually and in combination, on a database containing 1600
samples of different numerals and the results are compared with the
results of different existing methods.
Abstract: In this paper, we present a new learning algorithm for
anomaly based network intrusion detection using improved self
adaptive naïve Bayesian tree (NBTree), which induces a hybrid of
decision tree and naïve Bayesian classifier. The proposed approach
scales up the balance detections for different attack types and keeps
the false positives at acceptable level in intrusion detection. In
complex and dynamic large intrusion detection dataset, the detection
accuracy of naïve Bayesian classifier does not scale up as well as
decision tree. It has been successfully tested in other problem
domains that naïve Bayesian tree improves the classification rates in
large dataset. In naïve Bayesian tree nodes contain and split as
regular decision-trees, but the leaves contain naïve Bayesian
classifiers. The experimental results on KDD99 benchmark network
intrusion detection dataset demonstrate that this new approach scales
up the detection rates for different attack types and reduces false
positives in network intrusion detection.
Abstract: This paper deals with automatic sentence modality
recognition in French. In this work, only prosodic features are
considered. The sentences are recognized according to the three
following modalities: declarative, interrogative and exclamatory
sentences. This information will be used to animate a talking head for
deaf and hearing-impaired children. We first statistically study a real
radio corpus in order to assess the feasibility of the automatic
modeling of sentence types. Then, we test two sets of prosodic
features as well as two different classifiers and their combination. We
further focus our attention on questions recognition, as this modality
is certainly the most important one for the target application.
Abstract: In this paper we examine the use of global texture analysis based approaches for the purpose of Persian font recognition in machine-printed document images. Most existing methods for font recognition make use of local typographical features and connected component analysis. However derivation of such features is not an easy task. Gabor filters are appropriate tools for texture analysis and are motivated by human visual system. Here we consider document images as textures and use Gabor filter responses for identifying the fonts. The method is content independent and involves no local feature analysis. Two different classifiers Weighted Euclidean Distance and SVM are used for the purpose of classification. Experiments on seven different type faces and four font styles show average accuracy of 85% with WED and 82% with SVM classifier over typefaces
Abstract: Feature and model selection are in the center of
attention of many researches because of their impact on classifiers-
performance. Both selections are usually performed separately but
recent developments suggest using a combined GA-SVM approach to
perform them simultaneously. This approach improves the
performance of the classifier identifying the best subset of variables
and the optimal parameters- values. Although GA-SVM is an
effective method it is computationally expensive, thus a rough
method can be considered. The paper investigates a joined approach
of Genetic Algorithm and kernel matrix criteria to perform
simultaneously feature and model selection for SVM classification
problem. The purpose of this research is to improve the classification
performance of SVM through an efficient approach, the Kernel
Matrix Genetic Algorithm method (KMGA).
Abstract: Purpose of this work is to develop an automatic classification system that could be useful for radiologists in the breast cancer investigation. The software has been designed in the framework of the MAGIC-5 collaboration. In an automatic classification system the suspicious regions with high probability to include a lesion are extracted from the image as regions of interest (ROIs). Each ROI is characterized by some features based generally on morphological lesion differences. A study in the space features representation is made and some classifiers are tested to distinguish the pathological regions from the healthy ones. The results provided in terms of sensitivity and specificity will be presented through the ROC (Receiver Operating Characteristic) curves. In particular the best performances are obtained with the Neural Networks in comparison with the K-Nearest Neighbours and the Support Vector Machine: The Radial Basis Function supply the best results with 0.89 ± 0.01 of area under ROC curve but similar results are obtained with the Probabilistic Neural Network and a Multi Layer Perceptron.
Abstract: This paper discusses a method for improving accuracy
of fuzzy-rule-based classifiers using particle swarm optimization
(PSO). Two different fuzzy classifiers are considered and optimized.
The first classifier is based on Mamdani fuzzy inference system
(M_PSO fuzzy classifier). The second classifier is based on Takagi-
Sugeno fuzzy inference system (TS_PSO fuzzy classifier). The
parameters of the proposed fuzzy classifiers including premise
(antecedent) parameters, consequent parameters and structure of
fuzzy rules are optimized using PSO. Experimental results show that
higher classification accuracy can be obtained with a lower number
of fuzzy rules by using the proposed PSO fuzzy classifiers. The
performances of M_PSO and TS_PSO fuzzy classifiers are compared
to other fuzzy based classifiers
Abstract: The paper presents a multimodal approach for biometric authentication, based on multiple classifiers. The proposed solution uses a post-classification biometric fusion method in which the biometric data classifiers outputs are combined in order to improve the overall biometric system performance by decreasing the classification error rates. The paper shows also the biometric recognition task improvement by means of a carefully feature selection, as much as not all of the feature vectors components support the accuracy improvement.
Abstract: A recent neurospiking coding scheme for feature extraction from biosonar echoes of various plants is examined with avariety of stochastic classifiers. Feature vectors derived are employedin well-known stochastic classifiers, including nearest-neighborhood,single Gaussian and a Gaussian mixture with EM optimization.Classifiers' performances are evaluated by using cross-validation and bootstrapping techniques. It is shown that the various classifers perform equivalently and that the modified preprocessing configuration yields considerably improved results.
Abstract: This paper illustrates the use of a combined neural
network model for classification of electrocardiogram (ECG) beats.
We present a trainable neural network ensemble approach to develop
customized electrocardiogram beat classifier in an effort to further
improve the performance of ECG processing and to offer
individualized health care.
We process a three stage technique for detection of premature
ventricular contraction (PVC) from normal beats and other heart
diseases. This method includes a denoising, a feature extraction and a
classification. At first we investigate the application of stationary
wavelet transform (SWT) for noise reduction of the
electrocardiogram (ECG) signals. Then feature extraction module
extracts 10 ECG morphological features and one timing interval
feature. Then a number of multilayer perceptrons (MLPs) neural
networks with different topologies are designed.
The performance of the different combination methods as well as
the efficiency of the whole system is presented. Among them,
Stacked Generalization as a proposed trainable combined neural
network model possesses the highest recognition rate of around 95%.
Therefore, this network proves to be a suitable candidate in ECG
signal diagnosis systems. ECG samples attributing to the different
ECG beat types were extracted from the MIT-BIH arrhythmia
database for the study.
Abstract: There are two common methodologies to verify
signatures: the functional approach and the parametric approach. This
paper presents a new approach for dynamic handwritten signature
verification (HSV) using the Neural Network with verification by the
Conjugate Gradient Neural Network (NN). It is yet another avenue in
the approach to HSV that is found to produce excellent results when
compared with other methods of dynamic. Experimental results show
the system is insensitive to the order of base-classifiers and gets a
high verification ratio.
Abstract: This paper presents an effective traffic lights
recognition method at the daytime. First, Potential Traffic Lights
Detector (PTLD) use whole color source of YCbCr channel image and
make each binary image of green and red traffic lights. After PTLD
step, Shape Filter (SF) use to remove noise such as traffic sign, street
tree, vehicle, and building. At this time, noise removal properties
consist of information of blobs of binary image; length, area, area of
boundary box, etc. Finally, after an intermediate association step witch
goal is to define relevant candidates region from the previously
detected traffic lights, Adaptive Multi-class Classifier (AMC) is
executed. The classification method uses Haar-like feature and
Adaboost algorithm. For simulation, we are implemented through Intel
Core CPU with 2.80 GHz and 4 GB RAM and tested in the urban and
rural roads. Through the test, we are compared with our method and
standard object-recognition learning processes and proved that it
reached up to 94 % of detection rate which is better than the results
achieved with cascade classifiers. Computation time of our proposed
method is 15 ms.
Abstract: In recent years, a number of works proposing the
combination of multiple classifiers to produce a single
classification have been reported in remote sensing literature. The
resulting classifier, referred to as an ensemble classifier, is
generally found to be more accurate than any of the individual
classifiers making up the ensemble. As accuracy is the primary
concern, much of the research in the field of land cover
classification is focused on improving classification accuracy. This
study compares the performance of four ensemble approaches
(boosting, bagging, DECORATE and random subspace) with a
univariate decision tree as base classifier. Two training datasets,
one without ant noise and other with 20 percent noise was used to
judge the performance of different ensemble approaches. Results
with noise free data set suggest an improvement of about 4% in
classification accuracy with all ensemble approaches in
comparison to the results provided by univariate decision tree
classifier. Highest classification accuracy of 87.43% was achieved
by boosted decision tree. A comparison of results with noisy data
set suggests that bagging, DECORATE and random subspace
approaches works well with this data whereas the performance of
boosted decision tree degrades and a classification accuracy of
79.7% is achieved which is even lower than that is achieved (i.e.
80.02%) by using unboosted decision tree classifier.