classification

Open Access Article

1 - Search Space Reduction in Fingerprint Recognition Based on Block Orientation Field
S. Helfroush H. Ghassemian

20.1001.1.16823745.1385.4.8.1.7

Classification is the first essential step in every automatic fingerprint recognition system. Regarding to the time and expense of recognition process, it has the benefit of search space reduction. Conventional classification methods are based on visible fingerprint cla More

Classification is the first essential step in every automatic fingerprint recognition system. Regarding to the time and expense of recognition process, it has the benefit of search space reduction. Conventional classification methods are based on visible fingerprint classes. However, due to small number of these classes and nonuniform distribution of fingerprints among them, continuous classification scheme has been addressed. In this method, a similarity criterion is defined and a degree of likeness is assigned to the similarity of input fingerprint and each fingerprint in database. According to similarity criterion, matching of input fingerprint is begun first with the image in database that is more similar to input fingerprint. In this paper, a new similarity measuring method is proposed and used for continuous classification of fingerprints. The method is based on block orientation field. It is translation and rotation invariant and does not need core point existence and detection. Experimental results on FVC2000 database demonstrate the effectiveness of the proposed algorithm in search space reduction compared with the other methods. Manuscript profile

Open Access Article

2 - A Method for Automatic Printing Carpet Map Reading and Comparing to C-Means Clustering
Ahmad Izadipour E. Kabir

20.1001.1.16823745.1389.8.1.6.6

The subject of this paper is to read carpet pattern automatically by computer. This is composed of two steps: detection of vertical and horizontal lines in the pattern and color reduction. Color reduction is essential because of limitation of the number of colors that i More

The subject of this paper is to read carpet pattern automatically by computer. This is composed of two steps: detection of vertical and horizontal lines in the pattern and color reduction. Color reduction is essential because of limitation of the number of colors that is used in a carpet. To accomplish of this process, we must detect the grid lines on the carpet pattern automatically. These lines are two types: thin lines and thick lines. At the first stage, the distance between thin lines is obtained. Having the first thin line detected, the other thin lines are drawn using this distance. We use a Comb method for detection of thick lines. The major problem in line detection is lagging or leading of the lines due to the mismatch between sampling frequency of the scanner and image resolution. We compensate this distortion in various steps in our algorithm. In the second step, we want all the pixels in the same square, to have the same color. This is obtained by mapping colors to the best color in the palette. We propose three methods. In first method the user selects two selections per any colors. Palette is obtained from some processes in these selections. Those pixels that are in the middle of the squares are mapped to the palette. Then color histogram is computed. The color that has the maximum histogram value is assigned to the square. In order to decrease user’s interference, C-means clustering algorithm is used in two types. The centers of initial clusters are determined once with user’s interference and once randomly. Results of these three methods are compared. We tested our methods on 20 samples of carpet patterns, and the error rate was variable from 0.07% to 0.5% between samples. Manuscript profile

Open Access Article

3 - Training of MLP Neural Network for Data Classification by GSA Method
M. Dehbashian Seyed-Hamid Zahiri

20.1001.1.16823745.1389.8.4.5.1

Nowadays, several techniques have presented for data classification. One of these techniques is neural network that has attracted many interests. In this classifier, selection a suitable learning method is very important for training of the network. Error back propagati More

Nowadays, several techniques have presented for data classification. One of these techniques is neural network that has attracted many interests. In this classifier, selection a suitable learning method is very important for training of the network. Error back propagation is the most usual training method of neural networks that late convergence and stopping in local optimum points are its weakness. New approach in neural networks training is the usage of heuristic algorithms. This paper suggests a new learning method namely gravitational search algorithm (GSA) in training of neural network for data classification. GSA method is the latest and the most novel version of swarm intelligence optimization methods. This algorithm is inspired fby the law of Newtonian gravity and mass concept in nature. In this paper, a MLP neural network is trained for classification of five benchmark data set by GSA method. Also, the proposed method efficiency in training and testing of neural network compared with those of two training methods error back propagation and particle swarm optimization. Final results showed the GSA method extraordinary performance for data correct classification in most of cases. Also, in these experiments the GSA method produced stable results in all of cases. In addition, the run time of GSA method is shorter than that of the PSO. Manuscript profile

Open Access Article

4 - Gravity Oriented One-Class Classifier Based on Support Vector Data Descriptor
H. Ghafarian H. Sadoghi Yazdi Y. Allahyari

20.1001.1.16823745.1391.10.4.3.3

In this paper, a one-class classifier based on the Support Vector Data Descriptor (SVDD) is proposed. In SVDD, even outlier samples which are out of the decision boundary, are affecting the boundary. This increases the error of the classifier. In the proposed classifier More

In this paper, a one-class classifier based on the Support Vector Data Descriptor (SVDD) is proposed. In SVDD, even outlier samples which are out of the decision boundary, are affecting the boundary. This increases the error of the classifier. In the proposed classifier, decision boundary is determined by all of the samples through a gravity oriented approach. In this way, two classifier is proposed which in one of them knowledge about outliers are also considered. The optimization problem of the proposed method is convex and can be used with the kernel methods. Experiments on the behavior of the proposed classifier regarding changes of the parameters were done. Comparing results of experiments with results of SVDD and Density Induced SVDD shows that the proposed method can decrease the effects of outliers. Manuscript profile

Open Access Article

5 - Developing a New Version of Local Binary Patterns for Texture Classification
M. Pakdel M. H. Shakoor

20.1001.1.16823745.1392.11.2.3.6

Texture classification is one of the main steps in image processing and computer vision applications. Feature extraction is the first step of texture classification process which plays a main role. Many approaches have proposed to classify textures since now. Among them More

Texture classification is one of the main steps in image processing and computer vision applications. Feature extraction is the first step of texture classification process which plays a main role. Many approaches have proposed to classify textures since now. Among them, Local Binary Patterns and Modified Local Binary Patterns, because of simplicity and classification accuracy, have emerged as one of the most popular ones. The Local Binary Patterns have simple implementation, but with increase in the radius of neighborhood, computational complexity will be increased. Modified Local Binary Patterns assigns various labels to uniform textures and a unique label to all non-uniform ones. In this respect, the modified local binary pattern can't classify non uniform textures as well as uniform ones. In this paper a new version of Local Binary Pattern is proposed that has less computational complexity than Local Binary Patterns and more classification accuracy than Modified version. The proposed approach classifies non uniform textures as well as uniform ones. Also with change in the length of central gray level intervals, locality and globally of the features can be controlled. Classification accuracy on two standard datasets, Brodatz and Outex, indicates the efficiency of the proposed approach. Manuscript profile

Open Access Article

6 - A Secure Algorithm to Overcome Fingerprint Classification Problems
F. Mirzaei H. Ebrahimpour-Komleh M. Biglari

20.1001.1.16823745.1393.12.4.7.1

Fingerprint as a biometric has the most applications in verification and identification systems, because of its specific properties. In identification systems, input image is compared with all of images stored in the database. In huge databases, the comparison will take More

Fingerprint as a biometric has the most applications in verification and identification systems, because of its specific properties. In identification systems, input image is compared with all of images stored in the database. In huge databases, the comparison will take large amounts of time; Consider FBI databases, for instance. Image classification is one of the approved methods to increase the identification speed. Only one class is assigned to each fingerprint in tradition absolute classification. Various reasons like noise or lack of all the singularity points in captured region, cause the problem in determination of an absolute class for all the images. In this article, a new method based on probabilistic classification is presented. In the proposed approach, a set of classes are considered for each input image with a specific probability. These classes are searched in order of their probabilities priority in matching stage. Experiments on well-known FVC2002 database, exhibit the effect of probable classification clearly. Using only the second and third classes assigned by the proposed method, the identification system achieves about 18% increase in accuracy and 2-3 times speedup in compared to the traditional methods. Manuscript profile

Open Access Article

7 - Separating Bichromatic Point Sets by Right Triangles
Z. Moslehi A. Bagheri

20.1001.1.16823745.1394.13.4.9.0

Separating colored point sets is an interesting problem in computational geometry with application in machine learning and pattern recognition. In this problem, we are given a geometric shape C and two point sets P and Q of total size n as red and blue points, respectiv More

Separating colored point sets is an interesting problem in computational geometry with application in machine learning and pattern recognition. In this problem, we are given a geometric shape C and two point sets P and Q of total size n as red and blue points, respectively. Now, we must separate red and blue points by this shape such that all the blue points lie inside it and all the red points lie outside it. In the previous work, we have some algorithms for rectangle and wedge separability but we do not have any algorithm for separating by a triangle and separating by a triangle with a fixed angle such as right triangle. In this paper, we present an efficient algorithm for right triangle seprability. In this algorithm, we use sweep line technique and introduce some events and process them. So, we can report all separating right triangles in O(nlog n) time. Manuscript profile

Open Access Article

8 - A Novel Extended Mapping of Local Binary Pattern for Texture Classification
M. H. Shakoor M. H. Shakoor

20.1001.1.16823745.1395.14.3.11.7

Texture classification is one of the important branches of image processing. The main point of texture classification is feature extraction. Local Binary Pattern (LBP) is one of the important methods that are used for texture feature extraction. This method is widely us More

Texture classification is one of the important branches of image processing. The main point of texture classification is feature extraction. Local Binary Pattern (LBP) is one of the important methods that are used for texture feature extraction. This method is widely used because it has simple implementation and extracts high discriminative features from textures. Most of previous LBP methods used uniform patterns and only one feature is extracted from non-uniform patterns. In this paper, by extending non-uniform patterns a new mapping technique is proposed that extracts more discriminative features from non-uniform patterns. So in spite of almost all of the previous LBP methods, the proposed method extracts more discriminative features from non-uniform patterns and increases the classification accuracy of textures. The proposed method has all of the positive points of previous LBP variants. It is a rotation invariant and illumination invariant method and increase the classification accuracy. The implementation of proposed mapping on Outex dataset shows that proposed method can improve the accuracy of classifications significantly. Manuscript profile

Open Access Article

9 - A Novel Cascading Scheme to Improve Speed and Accuracy of a VMMR System
M. Biglari

In the last decade, many researches have been done on fine-grained recognition. The main category of the object is known in this problem and the goal is to determine the subcategory or fine-grained category. Vehicle Make and Model Recognition (VMMR) is a hard fine-grain More

In the last decade, many researches have been done on fine-grained recognition. The main category of the object is known in this problem and the goal is to determine the subcategory or fine-grained category. Vehicle Make and Model Recognition (VMMR) is a hard fine-grained classification problem, due to the large number of classes, substantial inner-class and small inter-class distance. Furthermore, improving system accuracy leads to increasing in processing time. As we can see the state-of-the-art machine vision tool like convolutional neural networks lacks in real-time processing time. In this paper, a method has been presented briefly for VMMR firstly. Secondly, a cascading scheme for improving both speed and accuracy of this VMMR system has been proposed. In order to eliminate extra processing cost, the proposed cascading scheme applies classifiers to the input image in a sequential manner. Some effective criterions for an efficient ordering of classifiers are proposed and finally a fusion of them is used in the cascade algorithm. For evaluation purposes, a new dataset with more than 5000 vehicles of 28 different makes and models has been collected. The experimental results on this dataset and comprehensive CompCars dataset show outstanding performance of our approach. Our cascading scheme results up to 80% increase in the system processing speed. Manuscript profile

Open Access Article

10 - A Novel Cascading Scheme to Improve Speed and Accuracy of a VMMR System
M. Biglari ali Soleimani H. Hassanpour

20.1001.1.16823745.1396.15.3.12.5

In the last decade, many researches have been done on fine-grained recognition. The main category of the object is known in this problem and the goal is to determine the subcategory or fine-grained category. Vehicle Make and Model Recognition (VMMR) is a hard fine-grain More

In the last decade, many researches have been done on fine-grained recognition. The main category of the object is known in this problem and the goal is to determine the subcategory or fine-grained category. Vehicle Make and Model Recognition (VMMR) is a hard fine-grained classification problem, due to the large number of classes, substantial inner-class and small inter-class distance. Furthermore, improving system accuracy leads to increasing in processing time. As we can see the state-of-the-art machine vision tool like convolutional neural networks lacks in real-time processing time. In this paper, a method has been presented briefly for VMMR firstly. Secondly, a cascading scheme for improving both speed and accuracy of this VMMR system has been proposed. In order to eliminate extra processing cost, the proposed cascading scheme applies classifiers to the input image in a sequential manner. Some effective criterions for an efficient ordering of classifiers are proposed and finally a fusion of them is used in the cascade algorithm. For evaluation purposes, a new dataset with more than 5000 vehicles of 28 different makes and models has been collected. The experimental results on this dataset and comprehensive CompCars dataset show outstanding performance of our approach. Our cascading scheme results up to 80% increase in the system processing speed. Manuscript profile

Open Access Article

11 - Adaptive Traffic Classification Algorithm for Live IPTV in EPON
M. Ahmadzadeh Bolghan Olia Mohammad Behdadfa M .R. Nourifard

20.1001.1.16823745.1396.15.4.17.2

In this paper, An IPTV packet classification algorithm is introduced with adaptive adjustment property and the objective of reducing packet loss ratio in Ethernet passive optical networks. The proposed algorithm, improves weight allocation for WRR scheduling using prope More

In this paper, An IPTV packet classification algorithm is introduced with adaptive adjustment property and the objective of reducing packet loss ratio in Ethernet passive optical networks. The proposed algorithm, improves weight allocation for WRR scheduling using proper classification and prioritization of arriving packets to OLT. Based on simulation results with NS2 simulator, the packet loss ratio of proposed algorithm, shows a 65% reduction compared to existing methods which leads to perceptible user quality of experience improvement. Manuscript profile

Open Access Article

12 - Attribute Reduction Based on Rough Set Theory by Soccer League Competition Algorithm
M. Abdolrazzagh-Nezhad Ali Adibiyan

20.1001.1.16823745.1399.18.3.10.4

Increasing the dimension of the databases have involved the attribute reduction as a critical issue in data mining that it searches to find a subset of attributes with the most effectiveness on the hidden patterns. In the current years, the rough set theory has been con More

Increasing the dimension of the databases have involved the attribute reduction as a critical issue in data mining that it searches to find a subset of attributes with the most effectiveness on the hidden patterns. In the current years, the rough set theory has been considered by researchers as one of the most effective and efficient tools to the reduction. In this paper, the soccer league competition algorithm is modified and adopted to solve the attribute reduction problem for the first time. The ability to escape the local optimal, the ability to use the information distributed by players in the search space, the rapid convergence to the optimal solutions, and the low algorithm’s parameters were the motivation of considering the algorithm in the current research. The proposed ideas to modify the algorithm consist of utilizing the total power of fixed and saved players in calculating the power of each team, considering the combination of continuous and discrete structures for each player, proposing a novel discretization method, providing a hydraulic analysis appropriate to the research problem for evaluating each player, designing correction in Imitation and Provocation operators based on the challenges in their original version. The proposed ideas are performed on small, medium and large data sets from UCI and the experimental results are compared with the state-of-the-art algorithms. This comparison shows that the competitive advantages of the proposed algorithm over the investigated algorithms. Manuscript profile

Open Access Article

13 - Cautious Classification of Hyper Rectangular, Hyper Circular, and Hyper Oval with a Maximum Symmetric Margin Relative to the Data Edge
Yahya Forghani M. Hejazi H. Sadoghi Yazdi

20.1001.1.16823745.1398.17.3.9.6

A robust classification model is a non-standard model for classifying learning based on an uncertain data set. An incautious model is said to have any meaningless answer to any classification model in its possible set of possible solutions. The optimal answer for a caut More

A robust classification model is a non-standard model for classifying learning based on an uncertain data set. An incautious model is said to have any meaningless answer to any classification model in its possible set of possible solutions. The optimal answer for a cautious robust classification model for a training data set may not be the hyper-page, in which case it will not be possible to classify the data at the test stage. In this paper, incautious robust classification models are introduced and their problems are investigated and then by changing the loss function of a robust classifier, a cautious robust classification model is presented to prevent incautious. The proposed cautious model is standardized and solutions are provided to reduce the training time and test time. In the experiments, the proposed model was compared with some incautious robust models to classification incomplete training data set, and complete definitive training data set. The results showed that in the incomplete data set, the proposed model had less training time and error rate than incautious models. Also, in the complete definitive data set, the proposed model training time and test time were less than incautious models. The results approved that adding caution to a robust classifier is efficient. Manuscript profile

Open Access Article

14 - A Traffic-Aware Packet Classification Method to Reduce Memory Accesses
Saeid Asadrooz Mohammad Nassiri M. A.

20.1001.1.16823745.1399.18.3.15.9

Packet classification plays a critical role in improving the performance of many network devices including routers, firewalls and intrusion detection systems. Due to the increasing number of classification rules, high traffic volume and high bandwidth network links, des More

Packet classification plays a critical role in improving the performance of many network devices including routers, firewalls and intrusion detection systems. Due to the increasing number of classification rules, high traffic volume and high bandwidth network links, designing an efficient packet classifier becomes more challenging. Packet classification algorithms that use static data structure do not consider the pattern of the incoming traffic in optimizing their search mechanism. Therefore, we use some statistical characteristics of the incoming traffic to propose a traffic aware data structure. Since most Internet traffic volume belong to long-live flows, the majority of the packets are matched to the rules in a few sub trees. To take the advantage of this feature, AVL tree data structure is served for storing classification rules where the upper and lower limits of the rule-set are used as nodes. Our evaluation have shown that with increasing the skewness of data packets, the average number of memory accesses are significantly decreased compared to the basic case. Finally, evaluation results show that the traffic-aware packet classification with high frequency rules can decrease more than 40% of the average number of memory accesses and consequently the lookup time. Manuscript profile

Open Access Article

15 - Improving the Architecture of Convolutional Neural Network for Classification of Images Corrupted by Impulse Noise
Mohammad Momeny M. Agha Sarram A. M. Latif R. Sheikhpour

20.1001.1.16823745.1398.17.4.10.9

Impulse noise is one the common noises which reduces the performance of convolutional neural networks (CNNs) in image classification. Preprocessing for removal of impulse noise is a costly process which may have a destructive effect on the training and validation of the More

Impulse noise is one the common noises which reduces the performance of convolutional neural networks (CNNs) in image classification. Preprocessing for removal of impulse noise is a costly process which may have a destructive effect on the training and validation of the convolutional neural networks due to insufficient improvement of noisy images. In this paper, a convolutional neural network is proposed which is robust to impulse noise. Proposed CNN classify images corrupted by impulse noise without any preprocessing for noise removal. A noise detection layer is placed at the beginning of the proposed CNN to prevent the processing of noisy values. The ILSVRC-2012 database is used to train the proposed CNN. Experimental results show that preventing the impact of impulse noise on the training process and classification of CNN can increase the accuracy and speed of the network training. The proposed CNN with error of 0.24 is better than other methods in classification of noisy image corrupted by impulse noise with 10% density. The time complexity of O(1) in the proposed CNN for robustness to noise indicates the superiority of the proposed CNN. Manuscript profile

Open Access Article

16 - Propose a Proper Algorithm for Incremental Learning Based on Fuzzy Least Square Twin Support Vector Machines
Javad Salimi Sartakhti Salman Goli

20.1001.1.16823745.1400.19.3.3.4

Support Vector machine is one of the most popular and efficient algorithms in machine learning. There are several versions of this algorithm, the latest of which is the fuzzy least squares twin support vector machines. On the other hand, in many machine learning applica More

Support Vector machine is one of the most popular and efficient algorithms in machine learning. There are several versions of this algorithm, the latest of which is the fuzzy least squares twin support vector machines. On the other hand, in many machine learning applications input data is continuously generated, which has made many traditional algorithms inefficient to deal with them. In this paper, for the first time, an incremental version of the fuzzy least squares twin support vector algorithm is presented. The proposed algorithmis represented in both online and quasi-online modes. To evaluate the accuracy and precision of the proposed algorithmfirst we run our algorithm on 6 datasets of the UCI repository. Results showthe proposed algorithm is more efficient than other algorithms (even non-incremental versions). In the second phase in the experiments, we consider an application of Internet of Things, and in particular in data related to daily activities which inherently are incremental. According to experimental results, the proposed algorithm has the best performance compared to other incremental algorithms. Manuscript profile

Open Access Article

17 - An Approximate Binary Tree-Based Solution to Speed Up the Search for the Nearest Neighbor in Big Data
Hosein Kalateh M. D.

20.1001.1.16823745.1401.20.3.1.9

Due to the increasing speed of information production and the need to convert information into knowledge, old machine learning methods are no longer responsive. When using classifications with the old machine learning methods, especially the use of inherently lazy class More

Due to the increasing speed of information production and the need to convert information into knowledge, old machine learning methods are no longer responsive. When using classifications with the old machine learning methods, especially the use of inherently lazy classifications such as the k-nearest neighbor (KNN) method, the operation of classifying large data sets is very slow. Nearest Neighborhood is a popular method of data classification due to its simplicity and practical accuracy. The proposed method is based on sorting the training data feature vectors in a binary search tree to expedite the classification of big data using the nearest neighbor method. This is done by finding the approximate two farthest local data in each tree node. These two data are used as a criterion for dividing the data in the current node into two groups. The data set in each node is assigned to the left and right child of the current node based on their similarity to the two data. The results of several experiments performed on different data sets from the UCI repository show a good degree of accuracy due to the low execution time of the proposed method. Manuscript profile

Open Access Article

18 - Iranian Dastgah Music Recognition Based on Notes Sequence Extraction and Use of LSTM Networks
سینا غضنفری پور M. Khademi Abbas Ebrahimi moghadam

20.1001.1.16823745.1401.20.2.6.2

Iranian "Dastgah" music classification by computer is a very interesting yet complex and challenging topic for those who are interested in Iranian Dastgah music. The aforementioned problem is important, firstly, due to its many applications in different areas such as co More

Iranian "Dastgah" music classification by computer is a very interesting yet complex and challenging topic for those who are interested in Iranian Dastgah music. The aforementioned problem is important, firstly, due to its many applications in different areas such as composing and teaching music, and secondly, because of the needs of ordinary people to computer to detect the Dastgah. This paper presents a method for recognition of the genre (Dastgah) and subgenre (sub-Dastgah) of Iranian music based on sequential note extraction, hierarchical classification, and the use of LSTM networks. In the proposed method, the music track is first classified into one of the three general categories. The first category includes only "Mahour" Dastgah, the second category includes "Shour" and "Nava", and the third category includes "Homayoun", "Segah" and "Chahargah". Then, for each category, depending on its type, a different number of classifiers are applied until one of the 6 Dastgah and 11 sub-Dastgah of Iranian music are recognized. This research is not limited to any particular style of playing or instruments, it is also not affected by neither the speed nor the techniques of player. The labeled tracks in the "Arg" database, which is created for this research, are solo. However, some of them are also played by percussion instruments (such as the Tombak) along with melodic instruments. The results show that recognition of 6 main Dastgah and 11 sub-Dastgah have been approved by an average accuracy of 74.5% and 66.35%, respectively, which is more promising compared to other few similar studies. Manuscript profile

Open Access Article

19 - Semi-Supervised Self-Training Classification Based on Neighborhood Construction
mona emadi jafar tanha Mohammadebrahim Shiri Mehdi Hosseinzadeh Aghdam

20.1001.1.16823745.1401.20.3.5.3

Using the unlabeled data in the semi-supervised learning can significantly improve the accuracy of supervised classification. But in some cases, it may dramatically reduce the accuracy of the classification. The reason of such degradation is incorrect labeling of unlabe More

Using the unlabeled data in the semi-supervised learning can significantly improve the accuracy of supervised classification. But in some cases, it may dramatically reduce the accuracy of the classification. The reason of such degradation is incorrect labeling of unlabeled data. In this article, we propose the method for high confidence labeling of unlabeled data. The base classifier in the proposed algorithm is the support vector machine. In this method, the labeling is performed only on the set of the unlabeled data that is closer to the decision boundary from the threshold. This data is called informative data. the adding informative data to the training set has a great effect to achieve the optimal decision boundary if the predicted label is correctly. The Epsilon- neighborhood Algorithm (DBSCAN) is used to discover the labeling structure in the data space. The comparative experiments on the UCI dataset show that the proposed method outperforms than some of the previous work to achieve greater accuracy of the self-training semi-supervised classification. Manuscript profile

Open Access Article

20 - Multi-Label Feature Selection Using a Hybrid Approach Based on the Particle Swarm Optimization Algorithm
َAzar Rafiei Parham Moradi Abdolbaghi Ghaderzadeh

20.1001.1.16823745.1401.20.4.7.7

Multi-label classification is one of the important issues in machine learning. The efficiency of multi-label classification algorithms decreases drastically with increasing problem dimensions. Feature selection is one of the main solutions for dimension reduction in mul More

Multi-label classification is one of the important issues in machine learning. The efficiency of multi-label classification algorithms decreases drastically with increasing problem dimensions. Feature selection is one of the main solutions for dimension reduction in multi-label problems. Multi-label feature selection is one of the NP solutions, and so far, a number of solutions based on collective intelligence and evolutionary algorithms have been proposed for it. Increasing the dimensions of the problem leads to an increase in the search space and consequently to a decrease in efficiency and also a decrease in the speed of convergence of these algorithms. In this paper, a hybrid collective intelligence solution based on a binary particle swarm optimization algorithm and local search strategy for multi-label feature selection is presented. To increase the speed of convergence, in the local search strategy, the features are divided into two categories based on the degree of extension and the degree of connection with the output of the problem. The first category consists of features that are very similar to the problem class and less similar to other features, and the second category is similar features and less related. Therefore, a local operator is added to the particle swarm optimization algorithm, which leads to the reduction of irrelevant features and extensions of each solution. Applying this operator leads to an increase in the convergence speed of the proposed algorithm compared to other algorithms presented in this field. The performance of the proposed method has been compared with the most well-known feature selection methods on different datasets. The results of the experiments showed that the proposed method has a good performance in terms of accuracy. Manuscript profile

Open Access Article

21 - Combination of Instance Selection and Data Augmentation Techniques for Imbalanced Data Classification
Parastoo Mohaghegh Samira Noferesti Mehri Rajaei

Mohaghegh, S. Noferesti*, and M. Rajaei Abstract: In the era of big data, automatic data analysis techniques such as data mining have been widely used for decision-making and have become very effective. Among data mining techniques, classification is a common method fo More

Mohaghegh, S. Noferesti*, and M. Rajaei Abstract: In the era of big data, automatic data analysis techniques such as data mining have been widely used for decision-making and have become very effective. Among data mining techniques, classification is a common method for decision making and prediction. Classification algorithms usually work well on balanced datasets. However, one of the challenges of the classification algorithms is how to correctly predicting the label of new samples based on learning on imbalanced datasets. In this type of dataset, the heterogeneous distribution of the data in different classes causes examples of the minority class to be ignored in the learning process, while this class is more important in some prediction problems. To deal with this issue, in this paper, an efficient method for balancing the imbalanced dataset is presented, which improves the accuracy of the machine learning algorithms to correct prediction of the class label of new samples. According to the evaluations, the proposed method has a better performance compared to other methods based on two common criteria in evaluating the classification of imbalanced datasets, namely "Balanced Accuracy" and "Specificity". Manuscript profile

Current Issue

Published Issues

Menu

Browse

List of Articles classification