SIGNAL AND DATA PROCESSING
Year:2019 | Volume:15 | Issue:4 (38) |Papers Count:11

Clustering is the process of division of a dataset into subsets that are called clusters, so that objects within a cluster are similar to each other and different from objects of the other clusters. So far, a lot of algorithms in different approaches have been created for the clustering. An effective choice (can combine) two or more of these algorithms for solving the clustering problem. Ensemble clustering combines results of existing clusterings to achieve better performance and higher accuracy. Instead of combining all of existing clusterings, recent decade researchers show, if only a set of clusterings is selected based on quality and diversity, the result of ensemble clustering would be more accurate. This paper proposes a new method for ensemble clustering based on quality and diversity. For this purpose, firstly first we need a lot of different base clusterings to combine them. Different base clusterings are generated by k-means algorithm with random k in each execution. After the generation of base clusterings, they are put into different groups according to their similarities using a new grouping method. So that clusterings which are similar to each other are put together in one group. In this step, we use normalized mutual information (NMI) or adjusted rand index (ARI) for computing similarities and dissimilarities between the base clustering. Then from each group, a best qualified clustering is selected via a voting based method. In this method, Cluster-validity-indices were used to measure the quality of clustering. So that all members of the group are evaluated by the Cluster-validity-indices. In each group, clustering that optimizes the most number of Cluster-validity-indices is selected. Finally, consensus functions combine all selected clustering. Consensus function is an algorithm for combining existing clusterings to produce final clusters. In this paper, three consensus functions including CSPA, MCLA, and HGPA have used for combining clustering. To evaluate proposed method, real datasets from UCI repository have used. In experiment section, the proposed method is compared with the well-known and powerful existing methods. Experimental results demonstrate that proposed algorithm has better performance and higher accuracy than previous works.

Software defects detection is one of the most important challenges of software development and it is the most prohibitive process in software development. The early detection of fault-prone modules helps software project managers to allocate the limited cost, time, and effort of developers for testing the defect-prone modules more intensively. In this paper, according to the importance of software defects detection, a method based on fuzzy sets and evolutionary algorithms is proposed. Due to the imbalanced nature of software defect detection datasets, benefits of fuzzy clustering algorithms were used to data sampling and more attention to the minority class. This method is a combined algorithm which, firstly has used fuzzy c-mean clustering as weighted bootstrap sampling. Weight of data (their membership’ s degrees) increases for minority class. In the next step, the subtractive clustering algorithm is applied to produce the classifier which was trained by produced data in the previous step. The binary genetic algorithm was utilized to select appropriate features. The results and also comparisons with eight popular methods in software defect detection literature, show an acceptable performance of the proposed method. The experiments were performed on ten real-world datasets with a wide range of data sizes and imbalance rates. Also T-test is used as the statistical significance test for pair wise comparison of our proposed method against the others. The final results of T-test are shown in tables for three performance measures (G-mean, AUC and Balanced) over various datasets. (As the obtained results apparently show our proposed method has the ability to improve three aforementioned performance criteria simultaneously). Some methods just have improved the G-mean measure while the AUC and Balance criteria have lower values than the others. Securing a high level of three performance measures simultaneously illustrates the ability of our proposed algorithm for handling the imbalance problem of software defects detection datasets.

In secure multiparty computation (SMC), a group of users jointly and securely computes a mathematical function on their private inputs, such that the privacy of their private inputs will be preserved. One of the widely used applications of SMC is the secure multiparty summation which securely computes the summation value of the users’ private inputs. In this paper, we consider a secure multiparty summation problem where each group member has m private inputs and wants to efficiently and securely computes the summation values of their corresponding inputs; in other words, users compute m summation values where the first value is the summation of users’ first private inputs, the second one is the summation of users’ second private inputs and so on. We propose an efficient and secure protocol in the semi honest model, called frequent-sum, which computes the desired values while preserving the privacy of users’ private inputs as well as the privacy of the summation results. Let be a set of n users and the private inputs of user is denoted as. The proposed frequent-sum protocol includes three phases: In the first phase, each user selects a random number, computes and publishes the vectors of components where each component of is of form. After it, computes the vector, such that each component is of form. In the second phase, users jointly and securely compute their AV-net (Anonymous Veto network) masks and the Burmester-Desmedt (BD) conference key. To do so, each user selects two random numbers and and publishes to the group. Then, computes and sends to the group. Then, each user is able to compute and; is the AV-net mask of and is the conference key. In the third phase, using the AV-net mask and the conference key, group members securely and collaboratively compute the summation of their random numbers, . To achieve this, each user broadcasts to the group, where is the AV-net mask of and is the ’ s portion of the conference key. Multiplying all s results in canceling the AV-net mask and getting the value of. Then each member is able to compute by the following Eq.: Now each user is able to compute by subtracting from each component of: It is shown that the proposed protocol is secure against collusion attack of at most users. In other words, the frequent-sum protocol is secure against partial collusion attack; only a full collusion (collusion of users) would break the privacy of the victim user, in this situation there is no reason for the victim user to join to such a group. The performance analysis shows that the proposed protocol is efficient in terms of the computation and communication costs, comparing with previous works. Also, the computation cost of the frequent-sum protocol is in-dependent of the number of inputs of each user which makes the protocol more efficient than the previous works. Table 1 compares the proposed protocol with previous works.

An automatic Number Plate Recognition (ANPR) is a popular topic in the field of image processing and is considered from different aspects, since early 90s. There are many challenges in this field, including; fast moving vehicles, different viewing angles and different distances from camera, complex and unpredictable backgrounds, poor quality images, existence of multiple plates in the scene, variable lighting conditions throughout the day, and so on. ANPR systems have many applications in today’ s traffic monitoring and toll-gate systems. In this paper, a real-time algorithm is designed and implemented for simultaneous detection and recognition of multiple number plates in video sequences. Already some papers on plate localization and recognition in still? images have been existed, however, they do not consider real time processing. While for the related applications, real-time detection and recognition of multiple plates on the scene is very important. Unlike methods with high computational complexity, we apply simple and effective techniques for being real-time. At first, background is modeled using Gaussian Mixture Model (GMM) and moving objects are determined. Then, plate candidate regions are found by vertical edge detection and horizontal projection. After that, license plates are localized and extracted by morphological operations and connected components analysis. When plates were are detected, their characters are separated with another algorithm. Finally a neural network is applied for character recognition. This system is implemented in C++ using OpenCV library. The average localization time per frame is 25 ms and total processing time, including localization and recognition, is 40 ms that can be used in real-time applications. The proposed method is evaluated on videos from highway cameras and the detection rate of 98. 79% and recognition rate of 97. 83% is obtained. Our real-time system can also recognize multiple plates of different types in each frame. Experimental results show that our method have higher speed and better recognition rate than previous works therefore it is suitable for real-time applications.

A probabilistic topic model assumes that documents are generated through a process involving topics and then tries to reverse this process, given the documents and extract topics. A topic is usually assumed to be a distribution over words. LDA is one of the first and most popular topic models introduced so far. In the document generation process assumed by LDA, each document is a distribution over topics and each word in the document is sampled from a chosen topic of that distribution. It assumes that a document is a bag of words and ignores the order of the words. Probabilistic topic models such as LDA which extract the topics based on documents-level word co-occurrences are not equipped to benefit from local word relationships. This problem is addressed by combining topics and n-grams, in models like Bigram Topic Model (BTM). BTM modifies the document generation process slightly by assuming that there are several different distributions of words for each topic, each of which correspond to a vocabulary word. Each word in a document is sampled from one of the distributions of its selected topic. The distribution is determined by its previous word. So BTM relies on exact word orders to extract local word relationships and thus is challenged by sparseness. Another way to solve the problem is to break each document into smaller parts for example paragraphs and use LDA on these parts to extract more local word relationships in these small parts. Again, we will be faced with sparseness and it is well-known that LDA does not work well on small documents. In this paper, a new probabilistic topic model is introduced which assumes a document is a set of overlapping windows but does not break the document into those parts and assumes the whole document as a single distribution over topics. Each window corresponds to a fixed number of words in the document. In the assumed generation process, we walk through windows and decide on the topic of their corresponding words. Topics are extracted based on words co-occurrences in the overlapping windows and the overlapping windows affect the process of document generation because; the topic of a word is considered in all the other windows overlapping on the word. On the other words, the proposed model encodes local word relationships without relying on exact word order or breaking the document into smaller parts. The model, however, takes the word order into account implicitly by assuming the windows are overlapped. The topics are still considered as distributions over words. The proposed model is evaluated based on its ability to extract coherent topics and its clustering performance on the 20 newsgroups dataset. The results show that the proposed model extracts more coherent topics and outperforms LDA and BTM in the application of document clustering.

Finding high-quality web pages is one of the most important tasks of search engines. The relevance between the documents found and the query searched depends on the user observation and increases the complexity of ranking algorithms. The other issue is that users often explore just the first 10 to 20 results while millions of pages related to a query may exist. So search engines have to use suitable algorithms with high performance to find the most relevant pages. The ranking section is an important part of search engines. Ranking is a process in which the web page quality is estimated by the search engine. There are two main methods for ranking web pages. In the first method, ranking is done based on the documents’ content (traditional rankings). Models, such as Boolean model, probability model and vector space model are used to rank documents based on their contents. In the second method, based on the graph, web connections and the importance of web pages, ranking process is performed. Based on researches on search engines, the majority of user queries is more than one term. For queries with more than one term, two models can be used. The first model assumes that query terms are independent of each other while the second model considers a location and order dependency between query terms. Experiments show that in the majority of queries there are dependencies between terms. One of the parameters that can specify dependencies between query terms is the distance between query terms in the document. In this paper, a new definition of distance based on Minimum Weighted Displacement Model (MWDM) of document terms to accommodate the query terms is presented. In the Minimum Weighted Displacement Model (MWDM), we call the minimum number of words moving a text to match the query term by space. In addition, because most of the ranking algorithms use the TF (Term Frequency) to score documents and for queries more than one term, there is no clear definition of these parameters; in this paper, according to the new distance concept, Phrase Frequency and Inverted Document Frequency are defined. Also, algorithms to calculate them are presented. The results of the proposed algorithm compared with multiple corresponding algorithms shows a favorable increase in average precision.

Chronic kidney failure is one of the most widespread diseases in Iran and the world. In general, the disease is common in high health indexes societies due to increased longevity. Treatment for chronic kidney failure is dialysis and kidney transplantation. Kidney transplantation is an appropriate and effective strategy for patients with End-Stage Renal Disease (ESRD), and it provides a better life and reduces mortality risk for patients. In contrast to many benefits that kidney transplantation has in terms of improving physical and mental health and the life’ s quality in kidney transplantation patients, it may be rejected because of host's immune response to the received kidney, and it consequences the need for another transplantation, or even death will have to. In fact, a patient that can survive for years with dialysis, he may lose his life with an inappropriate transplantation or be forced into high-risk surgical procedures. According to the above, the study of predicting the survival of kidney transplantation, its effective factors and providing a model for purposing of high prediction accuracy is essential. Studies in the field of survival of kidney transplantation include statistical studies, artificial intelligence and machine learning. In all of the studies in this feild, researchers have sought to identify a more effective set of features in survival of transplantation and the design of predictive models with higher accuracy and lower error rate. This study carried out on 756 kidney transplant patients with 21 features of Imam Reza and Fourth Shahid Merab hospital in Kermanshah from 2001 to 2012. Some features set to binary value and other features have real continuous values. Due to data are unbalance, which led to convergence of classification model to majority class, so over sampling and under sampling techniques has been used for achieving higher accuracy. To identify the more effective features on the survival of the kidney transplantation, the genetic meta-heuristic algorithm is used. For this purpose binary coding for each chromosome has been used; it is combining three single-point, two-point, and uniform operators to make better generations, better convergence and achieve higher accuracy rate. The genetic search algorithm plays a vital role in searching for such a space in a reasonable time because data search space is exponential. In fact, in balanced data, genetic algorithm determines the effective factors and the K-nearest neighbor model with precision of classification as the evaluator function was used to predict the five-year survival of the kidney transplantation. Based on the results of this study, in comparison to similar studies for prediction of survival transplanted kidney, the five-year survival rate of transplanted kidney was appropriate in these models. Also the effective factors in over sampling and under sampling methods with a precision of 96. 8% and 89. 2% are obtained respectively. in addition weight, donor and recipient age, pre-transplantation urea, pre-transplantation creatinine, hemoglobin before and after transplantation, donor gender, donor and recipient RH, primary illness, donor age up 30 and receipt age up 40 were identified as the effective features on kidney transplantation survival. Comparing the results of this study with previous studies shows the superiority of the proposed model from the point of view of the models' precision. In particular, balancing the data along the selection of optimal features leads to a high precision predictive model.

Keywords are the main focal points of interest within a text, which intends to represent the principal concepts outlined in the document. Determining the keywords using traditional methods is a time consuming process and requires specialized knowledge of the subject. For the purposes of indexing the vast expanse of electronic documents, it is important to automate the keyword extraction task. Since keywords structure is coherent, we focus on the relation between words. Most of previous methods in Persian are based on statistical relation between words and didn’ t consider the sense relations. However, by existing ambiguity in the meaning, using these statistic methods couldn’ t help in determining relations between words. Our method for extracting keywords is a supervised method which by using lexical chain of words, new features are extracted for each word. Using these features beside of statistic features could be more effective in a supervised system. We have tried to map the relations amongst word senses by using lexical chains. Therefore, in the proposed model, “ FarsNet” plays a key role in constructing the lexical chains. Lexical chain is created by using Galley and McKeown's algorithm that of course, some changes have been made to the algorithm. We used java version of hazm library to determine candidate words in the text. These words were identified by using POS tagging and Noun phrase chunking. Ten features are considered for each candidate word. Four features related to frequency and position of word in the text and the rest related to lexical chain of the word. After extracting the keywords by the classifier, post-processing performs for determining Two-word key phrases that were not obtained in the previous step. The dataset used in this research was chosen from among Persian scientific papers. We only used the title and abstract of these papers. The results depicted that using semantic relations, besides statistical features, would improve the overall performance of keyword extraction for papers. Also, the Naive Bayes classifier gives the best result among the investigated classifiers, of course, eliminating some of the features of the lexical chain improved its performance.

The Internet of Things (IoT), is a new concept that its emergence has caused ubiquity of sensors in the human life. All data are collected, processed, and transmitted by these sensors. As the number of sensors increases, the first challenge in establishing a secure connection is authentication between sensors. Anonymity, lightweight, and trust between entities are other main issues that should be considered. However, this challenge also requires some features so that the authentication is done properly. Anonymity, light weight and trust between entities are among the issues that need to be considered. In this study, we have evaluated the authentication protocols concerning the Internet of Things and analyzed the security vulnerabilities and limitations found in them. A new authentication protocol is also proposed using the hash function and logical operators, so that the sensors can use them as computationally limited entities. This protocol is performed in two phases and supports two types of intra-cluster and inter-cluster communication. The analysis of proposed protocol shows that security requirements have been met and the protocol is resistant against various attacks. In the end, confidentiality and authentication of the protocol are proved applying AVISPA tool and the veracity of the protocol using the BAN logic. Focusing on this issue, in this paper, we have evaluated the authentication protocols in the Internet of Things and analyzed their limitations and security vulnerabilities. Moreover, a new authentication protocol is presented which the anonymity is its main target. The hash function and logical operators are used not only to make the protocol lightweight but also to provide some computational resources for sensors. In compiling this protocol, we tried to take into account three main approaches to covering the true identifier, generating the session key, and the update process after the authentication process. As with most authentication protocols, this protocol is composed of two phases of registration and authentication that initially register entities in a trusted entity to be evaluated and authenticated at a later stage by the same entity. It is assumed that in the proposed protocol we have two types of entities; a weak entity and a strong entity. The poor availability of SNs has low computing power and strong entities of CH and HIoTS that can withstand high computational overhead and carry out heavy processing. We also consider strong entities in the proposed protocol as reliable entities since the main focus of this research is the relationship between SNs. On the other hand, given the authenticity of the sensors and the transfer of the key between them through these trusted entities, the authenticity of the sensors is confirmed, and the relationship between them is also reliable. This protocol supports two types of intra-cluster and inter-cluster communication. The analysis of the proposed protocol shows that security requirements such as untraceability, scalability, availability, etc. have been met and it is resistant against the various attacks like replay attack, eavesdropping attack.

Part of speech tagging (POS tagging) is an ongoing research in natural language processing (NLP) applications. The process of classifying words into their parts of speech and labeling them accordingly is known as part-of-speech tagging, POS-tagging, or simply tagging. Parts of speech are also known as word classes or lexical categories. The purpose of POS tagging is determining the grammatical category of the words in a sentence. Grammatical and syntactical features of words are determined based on these tags. The function of existing tagging methods depends on the corpus. As if the educational and test data are extracted from a corpus, the methods are well-functioning, or if the number of educational data is low, especially in probabilistic methods, the accuracy level also decreases. The words used in sentences are often vague. For example, the word 'Mahrami' can be a noun or an adjective. Existing ambiguity can be eliminated by using neighbor words and an appropriate tagging method. Methods in this domain are divided into several categories such as: based on memory [2], rule based methods [5], statistical [6], and neural network [7]. The precision of more of these methods is an average of 95% [1]. In the paper [13], using the TnT probabilistic tagging and smoothing and variations on the estimation of the three-words likelihood function, a tagging model has been created that has reached 96. 7% in total on the Penn Treebank and NEGRA entities. [14] Using the representation of the dependency network and extensive use of lexical features, such as the conditional continuity of the sequence of words, as well as the effective use of the foreground in the linear models of linear logarithms and fine-grained modeling of the unknown words, on the Penn Treebank WSJ model, 97. 24% accuracy is achieved. The first work in Farsi that has used the word neighborhoods and the similarity distribution between them. The accuracy of the system is 57. 5%. In [19], a Persian open source tagger called HunPoS was proposed. This tag uses the same TnT method based on the Hidden Markov model and a triple sequence of words, and 96. 9% has reached on the ''Bi Jen Khan'' corpus. In this paper a statistical based method is proposed for Persian POS tagging. The limitations of statistical methods are reduced by introducing a fuzzy network model, such that the model is able to estimate more reliable parameters with a small set of training data. In this method, normalization is done as a preprocessing step and then the frequency of each word is estimated as a fuzzy function with respect to the corresponding tag. Then the fuzzy network model is formed and the weight of each edge is determined by means of a neural network and a membership function. Eventually, after the construction of a fuzzy network model for a sentence, the Viterbi algorithm as s subset of Hidden Markov Model (HMM) algorithms is used to specify the most probable path in the network. The goal of this paper is to solve a challenge of probabilistic methods when the data is low and estimation made by these models is mistaken. The results of testing this method on ``Bi Jen Khan'' corpus verified that the proposed method has better performance than similar methods, like hidden Markov model, when fewer training examples are available. In this experiment, several times the data is divided into two groups of training and test with different sizes ascending. On the other hand, in the initial experiments, we reduced the train data size and, in subsequent experiments, increased its size and compared with the HMM algorithm. As shown in figure 4, the train and test set and are directly related to each other, as the error rate decreases with increasing the training set and vice versa. In tests, three criteria involving precision, recall and F1 have been used. In Table 4, the implementation of HMM models and a fuzzy network is compared with each other and the results are shown.