Abstract: Traffic control has been one of the most common and irritating problems since the time automobiles have hit the roads. Problems like traffic congestion have led to a significant time burden around the world and one significant solution to these problems can be the proper implementation of the Intelligent Transport System (ITS). It involves the integration of various tools like smart sensors, artificial intelligence, position technologies and mobile data services to manage traffic flow, reduce congestion and enhance driver's ability to avoid accidents during adverse weather. Road and traffic signs’ recognition is an emerging field of research in ITS. Classification problem of traffic signs needs to be solved as it is a major step in our journey towards building semi-autonomous/autonomous driving systems. The purpose of this work focuses on implementing an approach to solve the problem of traffic sign classification by developing a Convolutional Neural Network (CNN) classifier using the GTSRB (German Traffic Sign Recognition Benchmark) dataset. Rather than using hand-crafted features, our model addresses the concern of exploding huge parameters and data method augmentations. Our model achieved an accuracy of around 97.6% which is comparable to various state-of-the-art architectures.
Abstract: Choosing good features is an essential part of machine learning. Recent techniques aim to automate this process. For instance, feature learning intends to learn the transformation of raw data into a useful representation to machine learning tasks. In automatic audio classification tasks, this is interesting since the audio, usually complex information, needs to be transformed into a computationally convenient input to process. Another technique tries to generate features by searching a feature space. Genetic algorithms, for instance, have being used to generate audio features by combining or modifying them. We find this approach particularly interesting and, despite the undeniable advances of feature learning approaches, we wanted to take a step forward in the use of genetic algorithms to find audio features, combining them with more conventional methods, like PCA, and inserting search control mechanisms, such as constraints over a confusion matrix. This work presents the results obtained on particular audio classification problems.
Abstract: Feature Selection is significant in order to perform constructive classification in the area of cancer diagnosis. However, a large number of features compared to the number of samples makes the task of classification computationally very hard and prone to errors in microarray gene expression datasets. In this paper, we present an innovative method for selecting highly informative gene subsets of gene expression data that effectively classifies the cancer data into tumorous and non-tumorous. The hybrid gene selection technique comprises of combined Mutual Information and Fisher score to select informative genes. The gene selection is validated by classification using Support Vector Machine (SVM) which is a supervised learning algorithm capable of solving complex classification problems. The results obtained from improved Mutual Information and F-Score with SVM as a classifier has produced efficient results.
Abstract: Advances in spatial and spectral resolution of satellite
images have led to tremendous growth in large image databases. The
data we acquire through satellites, radars, and sensors consists of
important geographical information that can be used for remote
sensing applications such as region planning, disaster management.
Spatial data classification and object recognition are important tasks
for many applications. However, classifying objects and identifying
them manually from images is a difficult task. Object recognition is
often considered as a classification problem, this task can be
performed using machine-learning techniques. Despite of many
machine-learning algorithms, the classification is done using
supervised classifiers such as Support Vector Machines (SVM) as the
area of interest is known. We proposed a classification method,
which considers neighboring pixels in a region for feature extraction
and it evaluates classifications precisely according to neighboring
classes for semantic interpretation of region of interest (ROI). A
dataset has been created for training and testing purpose; we
generated the attributes by considering pixel intensity values and
mean values of reflectance. We demonstrated the benefits of using
knowledge discovery and data-mining techniques, which can be on
image data for accurate information extraction and classification from
high spatial resolution remote sensing imagery.
Abstract: In this paper, we propose the variational EM inference
algorithm for the multi-class Gaussian process classification model
that can be used in the field of human behavior recognition. This
algorithm can drive simultaneously both a posterior distribution of a
latent function and estimators of hyper-parameters in a Gaussian
process classification model with multiclass. Our algorithm is based
on the Laplace approximation (LA) technique and variational EM
framework. This is performed in two steps: called expectation and
maximization steps. First, in the expectation step, using the Bayesian
formula and LA technique, we derive approximately the posterior
distribution of the latent function indicating the possibility that each
observation belongs to a certain class in the Gaussian process
classification model. Second, in the maximization step, using a derived
posterior distribution of latent function, we compute the maximum
likelihood estimator for hyper-parameters of a covariance matrix
necessary to define prior distribution for latent function. These two
steps iteratively repeat until a convergence condition satisfies.
Moreover, we apply the proposed algorithm with human action
classification problem using a public database, namely, the KTH
human action data set. Experimental results reveal that the proposed
algorithm shows good performance on this data set.
Abstract: DNA Barcode provides good sources of needed
information to classify living species. The classification problem has
to be supported with reliable methods and algorithms. To analyze
species regions or entire genomes, it becomes necessary to use the
similarity sequence methods. A large set of sequences can be
simultaneously compared using Multiple Sequence Alignment which
is known to be NP-complete. However, all the used methods are still
computationally very expensive and require significant computational
infrastructure. Our goal is to build predictive models that are highly
accurate and interpretable. In fact, our method permits to avoid the
complex problem of form and structure in different classes of
organisms. The empirical data and their classification performances
are compared with other methods. Evenly, in this study, we present
our system which is consisted of three phases. The first one, is called
transformation, is composed of three sub steps; Electron-Ion
Interaction Pseudopotential (EIIP) for the codification of DNA
Barcodes, Fourier Transform and Power Spectrum Signal Processing.
Moreover, the second phase step is an approximation; it is
empowered by the use of Multi Library Wavelet Neural Networks
(MLWNN). Finally, the third one, is called the classification of DNA
Barcodes, is realized by applying the algorithm of hierarchical
classification.
Abstract: The problems arising from unbalanced data sets
generally appear in real world applications. Due to unequal class
distribution, many researchers have found that the performance of
existing classifiers tends to be biased towards the majority class. The
k-nearest neighbors’ nonparametric discriminant analysis is a method
that was proposed for classifying unbalanced classes with good
performance. In this study, the methods of discriminant analysis are
of interest in investigating misclassification error rates for classimbalanced
data of three diabetes risk groups. The purpose of this
study was to compare the classification performance between
parametric discriminant analysis and nonparametric discriminant
analysis in a three-class classification of class-imbalanced data of
diabetes risk groups. Data from a project maintaining healthy
conditions for 599 employees of a government hospital in Bangkok
were obtained for the classification problem. The employees were
divided into three diabetes risk groups: non-risk (90%), risk (5%),
and diabetic (5%). The original data including the variables of
diabetes risk group, age, gender, blood glucose, and BMI were
analyzed and bootstrapped for 50 and 100 samples, 599 observations
per sample, for additional estimation of the misclassification error
rate. Each data set was explored for the departure of multivariate
normality and the equality of covariance matrices of the three risk
groups. Both the original data and the bootstrap samples showed nonnormality
and unequal covariance matrices. The parametric linear
discriminant function, quadratic discriminant function, and the
nonparametric k-nearest neighbors’ discriminant function were
performed over 50 and 100 bootstrap samples and applied to the
original data. Searching the optimal classification rule, the choices of
prior probabilities were set up for both equal proportions (0.33: 0.33:
0.33) and unequal proportions of (0.90:0.05:0.05), (0.80: 0.10: 0.10)
and (0.70, 0.15, 0.15). The results from 50 and 100 bootstrap samples
indicated that the k-nearest neighbors approach when k=3 or k=4 and
the defined prior probabilities of non-risk: risk: diabetic as 0.90:
0.05:0.05 or 0.80:0.10:0.10 gave the smallest error rate of
misclassification. The k-nearest neighbors approach would be
suggested for classifying a three-class-imbalanced data of diabetes
risk groups.
Abstract: The paper presents the results of clusterization by
Kohonen self-organizing maps (SOM) applied for analysis of array of
Raman spectra of multi-component solutions of inorganic salts, for
determination of types of salts present in the solution. It is
demonstrated that use of SOM is a promising method for solution of
clusterization and classification problems in spectroscopy of multicomponent
objects, as attributing a pattern to some cluster may be
used for recognition of component composition of the object.
Abstract: Due to the fact that there exist only a small number of complex systems in artificial immune system (AIS) that work out nonlinear problems, nonlinear AIS approaches, among the well-known solution techniques, need to be developed. Gaussian function is usually used as similarity estimation in classification problems and pattern recognition. In this study, diagnosis of breast cancer, the second type of the most widespread cancer in women, was performed with different distance calculation functions that euclidean, gaussian and gaussian-euclidean hybrid function in the clonal selection model of classical AIS on Wisconsin Breast Cancer Dataset (WBCD), which was taken from the University of California, Irvine Machine-Learning Repository. We used 3-fold cross validation method to train and test the dataset. According to the results, the maximum test classification accuracy was reported as 97.35% by using of gaussian-euclidean hybrid function for fold-3. Also, mean of test classification accuracies for all of functions were obtained as 94.78%, 94.45% and 95.31% with use of euclidean, gaussian and gaussian-euclidean, respectively. With these results, gaussian-euclidean hybrid function seems to be a potential distance calculation method, and it may be considered as an alternative distance calculation method for hard nonlinear classification problems.
Abstract: Recent experimental evidences have shown that because
of a fast convergence and a nice accuracy, neural networks training
via extended kalman filter (EKF) method is widely applied. However,
as to an uncertainty of the system dynamics or modeling error, the
performance of the method is unreliable. In order to overcome this
problem in this paper, a new finite impulse response (FIR) filter based
learning algorithm is proposed to train radial basis function neural
networks (RBFN) for nonlinear function approximation. Compared
to the EKF training method, the proposed FIR filter training method
is more robust to those environmental conditions. Furthermore , the
number of centers will be considered since it affects the performance
of approximation.
Abstract: Bioinformatics methods for predicting the T cell
coreceptor usage from the array of membrane protein of HIV-1 are
investigated. In this study, we aim to propose an effective prediction
method for dealing with the three-class classification problem of
CXCR4 (X4), CCR5 (R5) and CCR5/CXCR4 (R5X4). We made
efforts in investigating the coreceptor prediction problem as follows: 1)
proposing a feature set of informative physicochemical properties
which is cooperated with SVM to achieve high prediction test
accuracy of 81.48%, compared with the existing method with
accuracy of 70.00%; 2) establishing a large up-to-date data set by
increasing the size from 159 to 1225 sequences to verify the proposed
prediction method where the mean test accuracy is 88.59%, and 3)
analyzing the set of 14 informative physicochemical properties to
further understand the characteristics of HIV-1coreceptors.
Abstract: Naïve Bayes classifiers are simple probabilistic
classifiers. Classification extracts patterns by using data file with a set
of labeled training examples and is currently one of the most
significant areas in data mining. However, Naïve Bayes assumes the
independence among the features. Structural learning among the
features thus helps in the classification problem. In this study, the use
of structural learning in Bayesian Network is proposed to be applied
where there are relationships between the features when using the
Naïve Bayes. The improvement in the classification using structural
learning is shown if there exist relationship between the features or
when they are not independent.
Abstract: In this paper, a new learning algorithm based on a
hybrid metaheuristic integrating Differential Evolution (DE) and
Reduced Variable Neighborhood Search (RVNS) is introduced to train
the classification method PROAFTN. To apply PROAFTN, values of
several parameters need to be determined prior to classification. These
parameters include boundaries of intervals and relative weights for
each attribute. Based on these requirements, the hybrid approach,
named DEPRO-RVNS, is presented in this study. In some cases, the
major problem when applying DE to some classification problems
was the premature convergence of some individuals to local optima.
To eliminate this shortcoming and to improve the exploration and
exploitation capabilities of DE, such individuals were set to iteratively
re-explored using RVNS. Based on the generated results on
both training and testing data, it is shown that the performance of
PROAFTN is significantly improved. Furthermore, the experimental
study shows that DEPRO-RVNS outperforms well-known machine
learning classifiers in a variety of problems.
Abstract: The issue of classifying objects into one of predefined
groups when the measured variables are mixed with different types
of variables has been part of interest among statisticians in many
years. Some methods for dealing with such situation have been
introduced that include parametric, semi-parametric and nonparametric
approaches. This paper attempts to discuss on a problem
in classifying a data when the number of measured mixed variables is
larger than the size of the sample. A propose idea that integrates a
dimensionality reduction technique via principal component analysis
and a discriminant function based on the location model is discussed.
The study aims in offering practitioners another potential tool in a
classification problem that is possible to be considered when the
observed variables are mixed and too large.
Abstract: In this paper we are interested in classification problems
with a performance constraint on error probability. In such
problems if the constraint cannot be satisfied, then a rejection option
is introduced. For binary labelled classification, a number of SVM
based methods with rejection option have been proposed over the
past few years. All of these methods use two thresholds on the SVM
output. However, in previous works, we have shown on synthetic data
that using thresholds on the output of the optimal SVM may lead to
poor results for classification tasks with performance constraint. In
this paper a new method for supervised classification with rejection
option is proposed. It consists in two different classifiers jointly
optimized to minimize the rejection probability subject to a given
constraint on error rate. This method uses a new kernel based linear
learning machine that we have recently presented. This learning
machine is characterized by its simplicity and high training speed
which makes the simultaneous optimization of the two classifiers
computationally reasonable. The proposed classification method with
rejection option is compared to a SVM based rejection method
proposed in recent literature. Experiments show the superiority of
the proposed method.
Abstract: Feature selection has recently been the subject of intensive research in data mining, specially for datasets with a large number of attributes. Recent work has shown that feature selection can have a positive effect on the performance of machine learning algorithms. The success of many learning algorithms in their attempts to construct models of data, hinges on the reliable identification of a small set of highly predictive attributes. The inclusion of irrelevant, redundant and noisy attributes in the model building process phase can result in poor predictive performance and increased computation. In this paper, a novel feature search procedure that utilizes the Ant Colony Optimization (ACO) is presented. The ACO is a metaheuristic inspired by the behavior of real ants in their search for the shortest paths to food sources. It looks for optimal solutions by considering both local heuristics and previous knowledge. When applied to two different classification problems, the proposed algorithm achieved very promising results.
Abstract: A new approach to promote the generalization ability
of neural networks is presented. It is based on the point of view of
fuzzy theory. This approach is implemented through shrinking or
magnifying the input vector, thereby reducing the difference between
training set and testing set. It is called “shrinking-magnifying
approach" (SMA). At the same time, a new algorithm; α-algorithm is
presented to find out the appropriate shrinking-magnifying-factor
(SMF) α and obtain better generalization ability of neural networks.
Quite a few simulation experiments serve to study the effect of SMA
and α-algorithm. The experiment results are discussed in detail, and
the function principle of SMA is analyzed in theory. The results of
experiments and analyses show that the new approach is not only
simpler and easier, but also is very effective to many neural networks
and many classification problems. In our experiments, the proportions
promoting the generalization ability of neural networks have even
reached 90%.
Abstract: Text Mining is around applying knowledge discovery
techniques to unstructured text is termed knowledge discovery in text
(KDT), or Text data mining or Text Mining. In decision tree
approach is most useful in classification problem. With this
technique, tree is constructed to model the classification process.
There are two basic steps in the technique: building the tree and
applying the tree to the database. This paper describes a proposed
C5.0 classifier that performs rulesets, cross validation and boosting
for original C5.0 in order to reduce the optimization of error ratio.
The feasibility and the benefits of the proposed approach are
demonstrated by means of medial data set like hypothyroid. It is
shown that, the performance of a classifier on the training cases from
which it was constructed gives a poor estimate by sampling or using a
separate test file, either way, the classifier is evaluated on cases that
were not used to build and evaluate the classifier are both are large. If
the cases in hypothyroid.data and hypothyroid.test were to be
shuffled and divided into a new 2772 case training set and a 1000
case test set, C5.0 might construct a different classifier with a lower
or higher error rate on the test cases. An important feature of see5 is
its ability to classifiers called rulesets. The ruleset has an error rate
0.5 % on the test cases. The standard errors of the means provide an
estimate of the variability of results. One way to get a more reliable
estimate of predictive is by f-fold –cross- validation. The error rate of
a classifier produced from all the cases is estimated as the ratio of the
total number of errors on the hold-out cases to the total number of
cases. The Boost option with x trials instructs See5 to construct up to
x classifiers in this manner. Trials over numerous datasets, large and
small, show that on average 10-classifier boosting reduces the error
rate for test cases by about 25%.
Abstract: Granular computing deals with representation of information in the form of some aggregates and related methods for transformation and analysis for problem solving. A granulation scheme based on clustering and Rough Set Theory is presented with focus on structured conceptualization of information has been presented in this paper. Experiments for the proposed method on four labeled data exhibit good result with reference to classification problem. The proposed granulation technique is semi-supervised imbibing global as well as local information granulation. To represent the results of the attribute oriented granulation a tree structure is proposed in this paper.
Abstract: This paper focuses on the data-driven generation
of fuzzy IF...THEN rules. The resulted fuzzy rule base can be
applied to build a classifier, a model used for prediction, or
it can be applied to form a decision support system. Among
the wide range of possible approaches, the decision tree and
the association rule based algorithms are overviewed, and two
new approaches are presented based on the a priori fuzzy
clustering based partitioning of the continuous input variables.
An application study is also presented, where the developed
methods are tested on the well known Wisconsin Breast Cancer
classification problem.