جستجوی مقالات مرتبط با کلیدواژه "support vector machine" در نشریات گروه "فناوری اطلاعات"

Many companies disrupt the economic system by violating their financial statements, which has led to a major economic crisis. There are various diagnostic strategies that are mostly human. These solutions have high costs for calculating and reviewing the financial statements of all companies, so we must look for a solution that can use data mining and automation to perform this diagnostic process. Of course, data mining methods have also been proposed for this case, each of which has advantages and disadvantages. The data mining methods presented so far have high computational overhead or low accuracy. However, in the method proposed in this research, the improved ID3 decision tree with Bayesian network and support vector machine has been used as a combined method. In this proposed method, to improve performance and accuracy, the rough set algorithm and hierarchical analysis are used to select effective features. The tree created in the proposed method has the lowest possible depth and therefore has a high velocity. The computational overhead of the proposed method is low due to the use of an optimal algorithm. The data used in the evaluation of 3-year data from 60 companies. In evaluating the proposed method, it is shown that the proposed method has an accuracy of 80%, which is considered to be high accuracy compared to similar methods. The time overhead in the proposed method is O(m.n) and the memory overhead is O(n) where m represents the size of the training set and n represents the feature set used in the training

Many companies disrupt the economic system by violating their financial statements, which has led to a major economic crisis. There are various diagnostic strategies that are mostly human. These solutions have high costs for calculating and reviewing the financial statements of all companies, so we must look for a solution that can use data mining and automation to perform this diagnostic process. Of course, data mining methods have also been proposed for this case, each of which has advantages and disadvantages. The data mining methods presented so far have high computational overhead or low accuracy. However, in the method proposed in this research, the improved ID3 decision tree with Bayesian network and support vector machine has been used as a combined method. In this proposed method, to improve performance and accuracy, the rough set algorithm and hierarchical analysis are used to select effective features. The tree created in the proposed method has the lowest possible depth and therefore has a high velocity. The computational overhead of the proposed method is low due to the use of an optimal algorithm. The data used in the evaluation of 3-year data from 60 companies. In evaluating the proposed method, it is shown that the proposed method has an accuracy of 80%, which is considered to be high accuracy compared to similar methods. The time overhead in the proposed method is O(m.n) and the memory overhead is O(n) where m represents the size of the training set and n represents the feature set used in the training

Water plays an important role in the growth and health of plants. The supply of this vital material affects not only agricultural yields, but also the quality and control of their physiological processes. Therefore, water scarcity has a negative effect on the growth of plants. In this research, in order to detect the plant's water requirement, a series of images of lilium plant under drought stress conditions were investigated to extract the color and morphological features and based on them, an intelligent system was designed to recognize the plant's water status. After analyzing the parameters extracted from the imagesusing statistical analysis at 5% probability level and the sequential forward feature ranking method, the most suitable features were selected to predict the moisture content of the plant; Furthermore, the classification was carried out using the support vector machine (SVM) with different kernels. Finally, it was shown that the classification accuracy for the linear kernels, sigmoeid, and RBF was with 6, 9, and 9 features, respectively, were 81.19, 81.04 and 83.12%, respectively. The results showed that the proposed system had the ability to determine the levels of stress and control the amount of water needed

Amyotrophic lateral sclerosis (ALS) is a neuromuscular disease, the most common disease of motor neurons. Since one of the most important early symptoms of the disease is the presence of movement disorders, the study of gait disorders has been the focus of many researchers. The aim of this study was to provide a suitable algorithm for the diagnosis of ALS. The data available in the Physionet database were used. They were recorded from 13 ALS patients and 16 healthy individuals. In this study, two methods of fusion have been employed to combine the information of the right and left foot signals, before feature extraction (signal-level fusion) and after feature extraction (feature-level fusion). We utilized the nonlinear features to characterize the gait signals of patients and healthy individuals, which includes: Energy-logarithm entropy, Shannon entropy, Higuchi fractal, and Katz fractal. Then, by performing the Wilcoxon statistical test on the extracted features, we tried to find significant differences between the groups. A support vector machine was used to separate ALS subjects from the normal group. The suggested algorithm has the ability to diagnose ALS with an average accuracy of 87%. The highest classification accuracy was obtained using the Katz feature, which is 100%. The proposed system based on fusion algorithms not only reduces the computational cost but also performs very well in providing separation rates. This framework could pave the way for the development of simple high-performance diagnostic systems in the future.

Question-answering system is a special type of information retrieval system. Considering a set of documents (such as Web or local set of documents), the system should answer the questions asked in the natural language form. Question-answering systems generally consist of question processing and analysis, keyword generation, information retrieval, answer extraction, and answer validation. In such systems, selecting proper answers for user question is an important task which influences performance of the whole system. An appropriate answer validation technique can increase the performance of the question-answering system. In this study, Support Vector Regression (SVR) is employed in order to perform answer validation. SVR eliminates the risk of getting stuck in local minima by reducing the operational risk. The proposed system is applied on some of the TREC and Wikipedia questions repositories. In order to evaluate performance of the proposed system, following measures are calculated: Mean Reciprocal Ranking (MRR) and F-measure. Based on experimental results, the proposed system achieved 81% MRR and 49.7% F-measure which shows higher performance than systems with no answer validation and also systems using neural network-based answer validation.

Breast cancer is the most common cancer among women and the existence of a precise and reliable system for the diagnosis of benign or malignant of this cancer is essential. Nowadays, using the results of needle aspiration cytology, data mining and machine learning techniques, early diagnosis of breast cancer can be done with greater accuracy. In this study, we propose a method consisting of two steps: in the first step, to eliminate the less important features, logistic regression has been used to select more important features. In the second step, the Support Vector Machine (SVM) classification algorithm has been used with three different kernel functions for the diagnosis of benign and malignant samples. To evaluate the performance of the proposed method, two data sets, WBCD and WDBC have been used with investigation of several metrics such as precision, the Area Under the ROC (AUC), true positive rate, false positive rate, accuracy and the F-measure. The results show that using the logistic regression method, it is possible to select the more efficient features, such that the proposed method reaches 98.69% in terms of classification accuracy.

Market forecasting, like stock's market, with high volume of transactions have an effect on researchers and investors and get their attention. Important issue factors in any investment decision are risk and turnover. Understanding market momentum gives the ability to predict future movements. The ability to predict in a market economy, enables to achieve a higher turnover by reducing risk and avoiding financial losses. News plays an important role in the process of evaluating the current stock price. The development of data mining methods, computational intelligence and machine learning algorithms have led to new models of prediction. phpCrawler is a php base content crawler that use of DomCralwer and Guzzle packages for storing web data. With this tool, the news releases are stored and categorized from 17 News Agency.    Then, by using text mining and Support Vector Machine with different kernel, predict stock price direction. In this research use 948990, news has been stored from 17 news agencies. More than 300,000 news regarding political and economic categories were used and stock prices of chemicals between November 2017 till March 2018 (123 trading days) were studied. The results show that by using the linear kernel Support Vector Machine algorithm, the prediction accuracy of average price movement reached to 83%. Using nonlinear kernel Support Vector Machine with poly kernel increased two percent prediction accuracy to 85% on average and other kernel had poorer prediction.

Internet and computer access have created opportunities for e-learning. Easier access to resources and freedom of action for users is one of the benefits of e-learning. However, e-learning is not as attractive and dynamic as traditional or face-to-face instruction, and in these systems the user's condition, such as learning rate and motivation, is not taken into account. Therefore, the developers of e-learning systems can help to solve the problems mentioned in these systems by considering the learning style and design of interactive user relationships. Automated identification of learning style not only increases the attractiveness of e-learning, but also increases the efficiency and motivation of learners in e-learning environments. Research shows that people differ in decision making, problem solving, and learning. Learning style makes people understand a story differently. For example, people with good visual memory prefer to present topics visually rather than orally. Applying a proper teaching method improves the learner's performance in the learning environment. Lack of attention to students' learning style reduces their motivation and interest in studying and engagement in educational courses. Students’ success is one of the prominent goals in the learning environments. In order to achieve this goal, paying attention to students’ learning style is essential. Being aware of students’ learning style helps to design an appropriate education method which improves student’s performance in the learning environments. In this paper, the aim is to create a model for automatic prediction of learning styles.

Therefore, two real datasets collected from an e-learning environment which consists of 202 electrical and computer engineering students. Behavioral features were extracted from users’ interaction with e-learning system and then learning styles were classified using twin support vector machine. Twin support vector machine is an extension of SVM which aims at generating two non-parallel hyperplanes. This classifier is not sensitive to imbalanced datasets and its training speed is fast.

In this study, increasing the attractiveness of e-learning is emphasized and the issue of automatic recognition of students' learning style has been investigated by MBTI model. Two data sets from the interaction of 202 electrical and computer engineering students with the Moodle e-learning system have been collected. The collected data set is very unbalanced, which has a negative effect on the accuracy of the categories. With this in mind, the twin support vector machine uses the least squares as a binder. The distinctive feature of this category is the low sensitivity to data balance and very high speed. The results show that the proposed method, despite the inconsistency of the data, has performed very well in the classification of students' learning style and accurately recognizes 95% of learning styles.

Due to the excellent performance of the proposed method, a new component can be added to e-learning systems such as Moodle by identifying the learning style, content and appropriate teaching method for the learner. Future research could also gather more data from an e-learning environment and categorize learning styles with cognitive characteristics from the learner.