In Optical Character Recognition (OCR), state-of-the-art algorithms are applied to the same databases to compare performance and cost. Various benchmark databases have been created recently in Persian script to facilitate OCR application development. Unfortunately, there is a lack of coherent categorization and systematic identification in Persian and other languages about how to choose the databases to provide a suitable platform for evaluation. This article provides an analytical framework called DB-QM (DataBase Qualitative Measurement) to achieve a macro vision for assessing Persian OCR databases. In our proposed framework, three components are available: First, a categorization of Persian databases is proposed. Therefore, the databases are considered from their content point of view. Second, several quantitative and qualitative evaluation criteria are introduced and categorized based on the nature of databases. Finally, a discussion about the strengths and weaknesses of databases is made on the proposed criteria. It concerns a comparison among databases and the critical points about how to select one for testing algorithms. In addition, the main challenges around database improvement have been taken into account. Our analytical discussion not only clarifies the superiority of one database to another but provides a diverse discipline on how to use the appropriate databases. Also, critical challenges for the enhancement of new databases are highlighted.

The cost of software testing could be reduced if faulty entities were identified prior to the testing phase, which is possible with software fault prediction (SFP). In most SFP models, machine learning (ML) methods are used, and one aspect of improving prediction accuracy with these methods is tuning their control parameters . However, parameter tuning has not been addressed properly in the field of software analytics , and the conventional methods (such as basic Differential Evolution, Random Search, and Grid Search) are either not up-to-date, or suffer from shortcomings, such as the inability to benefit from prior experience, or are overly expensive. This study aims to examine and propose parameter tuners, called DEPTs, based on different variants of Differential Evolution for SFP with the Swift-Finalize strategy (to reduce runtime), which in addition to being up-to-date, have overcome many of the challenges associated with common methods. An experimental framework was developed to compare DEPTs with three widely used parameter tuners, applied to four common data miners, on 10 open-source projects, and to evaluate the performance of DEPTs, we used eight performance measures . According to our results, the three tuners out of five DEPTs improved prediction accuracy in more than 70% of tuned cases, and occasionally, they exceeded benchmark methods by over 10% in case of G-measure. The DEPTs took reasonable amounts of time to tune parameters for SFP as well.

In most traditional machine learning algorithms, the training and testing datasets have identical distributions and feature spaces. However, these assumptions have not held in many real applications. Although transfer learning methods have been invented to fill this gap, they introduce new challenges as negative transfers (NTs). Most previous research considered NT a significant problem, but they pay less attention to solving it. This study will propose a transductive learning algorithm based on cellular learning automata (CLA) to alleviate the NT issue. Two famous learning automata (LA) entitled estimators are applied as estimator CLA in the proposed algorithms. A couple of new decision criteria called merit and and attitude parameters are introduced to CLA to limit NT. The proposed algorithms are applied to standard LA environments. The experiments show that the proposed algorithm leads to higher accuracy and less NT results.

A sudden fall accident is the main concern for the elderly and disabled people. Automatic detection of the falls from video sequences is an assistive technology for surveillance systems. In this study, a three-stage framework was presented and implemented based on the combination of the data from multiple cameras to address the challenges of occlusion and visibility. In the first stage, the number of used cameras was specified. In the second stage, each camera was decided locally based on its data about the fall incident. In the third and final stage, the aggregation function was used to combine the single camera’s decision considering the coverage rate coefficient of the used cameras. Experiments on the multiple-camera fall dataset demonstrated that our method is comparable to other state-of-the-art methods.

Summary The prediction of software artifacts on defect‐prone (DP) or non‐defect‐prone (NDP) classes during the testing phase helps minimize software business costs, which is a classification task in software defect prediction (SDP) field. Machine learning methods are helpful for the task, although they face the challenge of data imbalance distribution. The challenge leads to serious misclassification of artifacts, which will disrupt the predictor's performance. The previously developed stacking ensemble methods do not consider the cost issue to handle the class imbalance problem (CIP) over the training dataset in the SDP field. To bridge this research gap, in the cost‐sensitive stacked generalization (CSSG) approach, we try to combine the staking ensemble learning method with cost‐sensitive learning (CSL) since the CSL purpose is to reduce misclassification costs. In the cost‐sensitive stacked generalization (CSSG) approach, logistic regression (LR) and extremely randomized trees classifiers in cases of CSL and cost‐insensitive are used as a final classifier of stacking scheme. To evaluate the performance of CSSG, we use six performance measures. Several experiments are carried out to compare the CSSG with some cost‐sensitive ensemble methods on 15 benchmark datasets with different imbalance levels. The results indicate that the CSSG can be an effective solution to the CIP than other compared methods.

Drug repurposing is a vital function in pharmaceutical fields and has gained popularity in recent years in both the pharmaceutical industry and research community. It refers to the process of discovering new uses and indications for existing or failed drugs. It is cost-effective and reliable in contrast to experimental drug discovery, which is a costly, time-consuming, and risky process and limited to a relatively small number of targets. Accordingly, a plethora of computational methodologies have been propounded to repurpose drugs on a large scale by utilizing available high throughput data. The available literature, however, lacks a contemporary and comprehensive analysis of the current computational drug repurposing methodologies. In this paper, we presented a systematic analysis of computational drug repurposing which consists of three main sections: Initially, we categorize the computational drug repurposing methods based on their technical approach and artificial intelligence perspective and discuss the strengths and weaknesses of various methods. Secondly, some general criteria are recommended to analyze our proposed categorization. In the third and final section, a qualitative comparison is made between each approach which is a guide to understanding their preference to one another. Further, this systematic analysis can help in the efficient selection and improvement of drug repurposing techniques based on the nature of computational methods implemented on biological resources.

Today, knowledge graphs (KGs) are growing by enrichment and refinement methods. The enrichment and refinement can be gained using the correction and completion of the KG. The studies of the KG completion are rich, but less attention has been paid to the methods of the KG error correction. The correction methods are divided into embedding and nonembedding methods. Embedding correction methods have been recently introduced in which a KG is embedded into a vector space. Also, existing correction approaches focused on the recognition of the three types of errors, the outliers, inconsistencies and erroneous relations. One of the challenges is that most outlier correction methods can recognize only numeric outlier entities by nonembedding methods. On the other hand, inconsistency errors are recognized during the knowledge extraction step and existing methods of this field do not pay attention to the recognition of these errors as post-correction by embedding methods. Also, to correct erroneous relations, new embedding techniques have not been used. Since the errors of a KG are variant and there is no method to cover all of them, a new general correction method is proposed in this paper. This method is called correction tower in which these three error types are corrected in three trays. In this correction tower, a new configuration will be suggested to solve the above challenges. For this aim, a new embedding method is proposed for each tray. Finally, the evaluation results show that the proposed correction tower can improve the KG error correction methods and proposed configuration can outperform previous results.

In this study, the authors present a new feature extraction method for handwritten Arabic/Persian language word recognition. This feature is based on the angle, number, location, and size of straight lines which represents geometric and quantitative attributes of a word. At first, word image is broken into an m × n window and straight lines are extracted from each window. Then, the proposed features are taken from these lines and combined together. Finally, the features of the images are used for training and testing support vector machine classifier. The proposed method is tested on three datasets: IBN‐SINA and IFN/ENIT for Arabic words and Iran‐cities for Persian words recognition. Recognition accuracy of the proposed method is about 67.47, 86.22 and 80.78% for the Iran‐cities, IBN‐SINA and IFN/ENIT Arabic dataset, respectively, which is better than state‐of‐the‐art methods.

In this paper, an improved imperialist competitive algorithm called I-ICA is proposed for solving constrained optimisation problems. In I-ICA, the use of differential evolution (DE)/rand/1 mutation operator at assimilation step enhances the population diversity. Also, the binomial crossover operator improves the speed of convergence to optimal solution by distributing good information among solutions. Furthermore, Deb’s rules were applied for handling constrains. Examinations were done on 24 well-known functions and 5 engineering design problems. The comparison of I-ICA with basic ICA and 22 state-of-the-art algorithms shows I-ICA’s superiority in terms of the rate of convergence and quality of the obtained solution.

Community detection is one of the main challenges in social network analysis. Since the issue of community detection is considered as a NP-hard problem, Evolutionary algorithms have been used as one of the most effective approaches. In this paper, a multi-objective particle swarm optimization algor ithm and its extended versions are proposed. The aforementioned algorithm uses an opposition-based method for producing an initial swarm. It optimizes two objective functions at the same time which represents a partition of the network as well as using a mutation operator for handling the problem in high dimensions. The performances of the proposed algorithm and its extended versions have been evaluated on real networks. The result represented the efficiency of proposed methods. Also an optimum value is suggested for aforementioned algorithm that can be said with less complexity and calculations, proposed algorithm achieved the acceptable amount of accuracy. The remarkable thing is the better performance of algorithm as the size of social network grows.

Human activity recognition in Ambient Assisted Living (AAL) is an important application in health care systems and allows us to track regular activities or even predict these activities in order to monitor healthcare and find changes in patterns and lifestyles.A review of the literature reveals various approaches to discovering and recognizing human activities.The presence of a vast number of activity recognition issues and approaches has made it difficult to make adequate comparisons and accurate assessment.Introducing the five basic components of activity recognition in the smart homes as a famous environment to remote monitoring of patients and independent living for elderly, the present paper proposes SARF framework to classify each of activity recognition approaches and then it is evaluated based on the proposed classification by some proposed measures.Using SARF proposed framework can play an effective role in selecting the appropriate method for human activity recognition in smart homes and beneficial in analysis and evaluation of different methods for various challenges in this field.

Abstract Nowadays, because of the undeniable impact of the Internet on all aspects of human life, security preserving has received more attention. To reach an acceptable level of security, Completely Automatic Public Turing test to tell Computer and Human Apart or simply CAPTCHA as a security preserving tool has been tailored for situations that need to prevent bots from doing a specific action; for example, signing up and downloading. Simultaneously, it should be also designed in such a way to allow humans to perform the same action. Despite its advantageous applications, there are several important issues such as security, usability, and accessibility that make its use controversial. In this paper, attempts were made to do a comprehensive review on various aspects and state‐of‐the‐art of CAPTCHA in general and its alternatives in particular to help researchers easily focus on specific issues for the sake of proposing new solutions and ideas. Regarding the advancement in CAPTCHA development, new classifications were proposed to categorize different variations of CAPTCHAs and their problems and then compare them. Moreover, different types of CAPTCHAs' alternatives were classified and evaluated by introducing several proposed measures. This evaluation could come in handy for future studies that aim to develop new techniques for overcoming current deficiencies. Copyright © 2014 John Wiley & Sons, Ltd.

In recent years, many NLP systems and tasks are developed using machine learning methods. In order to achieve the best performance, these systems are generally trained on a large human annotated corpus. Since annotating such corpora is a very expensi

Text categorization is one of the fundamental tasks in text mining. Classical supervised methods need lot of labeled data to train a classifier. Since assigning labels to the large amount of data is very costly and time consuming, it is useful to use data sets without labels. So many different semi -supervised learning methods have been studied recently. Among these semi-supervised methods, self-training is one of the important learning algorithms that classifies unlabeled samples with small amount of labeled ones and adds the most confident samples to the training set. In this paper, dynamic weighting beside majority vote approach is applied to classify the unlabeled data to reliable and unreliable classes. Then, the reliable data are added to the training set and the remaining data including unreliable data are classified in iterative process. We tested this method on the extracted features of ten common Reuter-21578 classes. Experimental result indicates that proposed method improves the classification performance and it's effective.