Search

As a important part of teaching, Traditional paper analysis algorithm is complicated and not comprehensive enough. This paper according to the indicators of education measurement and improves the traditional calculation models, the system assesses the perfect calculation model, The teacher also can get the true and fast reaction to students' knowledge and the quality of the paper. This method thereby provides a convincing basis for the improvement on the future teaching and increase the effectiveness in teaching

In this paper, several data mining techniques were discussed and analyzed in order to achieve the objective of human daily activities recognition based on a continuous sensing data set. The data mining techniques of decision tree, Naïve Bayes and Neural Network were successfully applied to the data set. The paper also proposed an idea of combining the Neural Network with the Decision Tree, the result shows that it works much better than the typical Neural Network and the typical Decision Tree model.

A Chinese web-page classification algorithm based on SVM including the important aspects of text preprocessing, feature selection and multiple-Classification algorithm. In this paper, based on the analyses of features of Web documents, this paper does research the approach of classification in Support Vector Machine (SVM) and select of Kernel function. Furthermore, a web-page classification model and algorithm that is based on Binary Tree SVM is presented. The experiments show that it not only reduces the size of train set, but also has very high training efficiency. Its precision and recall are better.

the accurate data mining problem is studied in this paper. With the increasing of data attributes, degree of complexity of the data storage is also increased, resulting in that in data mining process, the complexity of computation is too high, reducing the convergence of the data mining method, thereby reducing the efficiency of data mining. To this end, this paper presents a data mining method based on association rules algorithm. The data is made simplified processing, to obtain the association rules between data which provides the basis for data mining. According to the association rules between the data, the data in line with the minimum support degree is calculated, to achieve accurate data mining. Experimental results show that the proposed algorithm for data mining, can improve mining efficiency, and achieve the desired results.

With the rapid development of network, the security problem of network becomes an issue which has been paid more and more attentions to. Among so many methods of intrusion prevention, data mining is a very effective one. The FP-growth algorithm is the most widely used algorithm for mining frequent item-sets, which is also an algorithm for mining association rules without candidate set. However, the FP-growth algorithm needs large memory when mining large database,and its running speed is slow. In order to overcome these problems, based on the FP-growth algorithm, this paper proposed an optimized algorithm. This paper compared the new algorithm with the previous one based on intrusion prevention model for campus network by experiments. Based on Experiments, we can draw the conclusion that, mining association rules by using the improved FP-growth algorithm can effectively detect the users’ behavior pattern, historical pattern and the current model to calculate the similarity of users, and provides the possibility to accurately judge the user behavior.

The goal of the presented article is a description of an inductive method of knowledge structure creating, which may be applied in a system of an autonomic vehicle control. The paper deals with the relatively simple task of industrial vehicles control, which operate in a specific environment. The information about the environment is achieved from camera mounted on the vehicle. An analysis of the image leads to the construction the knowledge in a form of structure of concepts (i.e. classes) and objects. The proposed approach is illustrated by simple example. The paper describes crucial problems of the proposed knowledge creation, especially concerning initial assumptions about visible phenomena and problems of the knowledge updating (learning). The article attempts to formulate conditions determining a successful usage of the proposed methodology. The main condition is building an efficient methods of searching the proper concept structure.

UCT (Upper confidential bounds on Trees) has been applied quite well as a selection approach in MCTS(Monte Carlo Tree Search) in imperfect information games like poker. By using risk dominance as complementary part of decision method besides payoff dominance, opponent strategies is better characterized as their risk factors, like bluff actions in Texas Hold’em Poker . In this paper, estimation method about the influence of risk factors on computing game equilibrium is provided. A novel algorithm, UCT-risk is proposed as modification about UCT algorithm basing on risk estimation methods. To verify the performance of new algorithm, Texas Hold’em, a popular test-bed for AI research is chosen as the experiment platform. The Agent adopted UCT-risk algorithm performs as well or better as the best previous approaches in experiments. And also it is applied in a poker agent named HITSZ_CS_13 in the 2013 AAAI Computer Poker Competition, which confirms the effectiveness of the UCT-risk provided in this paper.

Through the software defect prediction can effectively guide the rational distribution of software system development resources, so as to improve the quality of software and software reliability. In order to fully utilize the existing historical data to guide the software development of existing software system development, this paper based on an improved classification and regression tree (Classification and Regression, CART) algorithm software defect prediction models. The paper first principal component analysis of the data predicted correlation dimension (Principle Component Analysis, PCA) between data and reduce the data, and configured according to the theory and optimized CART decision tree algorithm, existing software defect prediction system, and with traditional defect prediction method, the experimental results show that the proposed prediction model has higher prediction accuracy and stability.

Intelligent Medical Systems currently has been used in most advanced hospitals worldwide, and most of theses systems have the function module of guiding-patient-diagnosis, which makes the patient use this feature to get a preliminary treatment opinion based on their patients condition. In this paper, the way of enhancing the effect of guiding-patient-diagnosis with boosting decision tree algorithm is researched, and the paper describes the specific applying process of the algorithm. Finally, boosting decision tree algorithm prove to improve the efficiency of intelligent guiding function, making intelligent guiding diagnosis function accurate and reliable, which save the health care resources and the time of doctors and patients.

In multimodal classification, we look for a set of strategies for mining and exploiting the most informative modalities for a given situation. These strategies are computations performed by the algorithms. In this paper, we propose to consider strategies as advice given to an algorithm by “expert.” There can be several classification strategies. Each strategy makes different assumptions regarding the fidelity of a sensor modality and uses different data to arrive at its estimates. Each strategy may place different trust in a sensor at different times, and each may be better in different situations. In this paper, we introduce a novel algorithm for combining expert strategies to achieve robust classification performance in a multimodal setting. We provide experimental results using real world examples to demonstrate the efficacy of the proposed algorithm.

BP neural network in character recognition, pattern classification, text and voice conversion, image compression, decision support and so on aspects has the widespread application, in view of the problems existing in the actual application, this paper researches learning algorithm and software implementation. Learning algorithm studies include three aspects, illustrates the basic thoughts of the BP algorithm, designed the three layers BP network structure, the mathematical model for the accurate description of algorithm. Software implementation studies include two aspects, the network model of all neurons become linked list structure and storage structure is designed, the design of the software process and will implement the process into four steps. BP algorithm of the software implementation is a basic work for the application of BP neural network, using the research results of this paper, the user can easily neural network design and simulation.

This paper describes the basic models of network security state evaluation system and concentrates on researching the situation assessment method with stochastic model. In the paper, which makes use of the Hidden Semi-Markov Model (HsMM), tries to simulate the operation of network system. The alert statistics, deriving from network defense system, is used as data sources to realize the evaluation of network security situation. HsMM modifies the HMM model concerning the hypothesis of some state-duration time in relation to exponential distribution, which coincides the description of the network systems operation in the real world. The experimental results imply that HsMM is an ideal security evaluation method.

Design automation is always one of research focuses in the field of industry practices. With the development of computer science and technology, speech recognition technology attracts increasing researchers, and it is widely applied to industrial production. Around design automation, we propose a geometry modeling method through design intent using speech recognition, which is an automatic and efficient geometry modeling method. This paper discusses the speech recognition for design intent of geometry modeling using Hidden Markov Model, HMM. In this paper, firstly, the research framework is explained in detail; Secondly, analysis on the related theory and technology about the speech recognition are conducted, including speech signal collection, technology of data processing, technology of training and recognition of model, etc.; Thirdly, some experiments are carried out to validate above discussions; Finally, recognition efficiency comparisons between HMM and DTW are discussed. Experiments show that using HMM for the speech recognition of design intent is feasible and reasonable.

The algorithm of frequent-term based text clustering (FTC) can be applied to news topic clustering system, in order to help users locate interested topics and articles quickly. But it is difficult to set support threshold for mining association rules. This paper tries to build a word belief network, which satisfies basic rules of small worlds. So we can improve FTC algorithm with characteristics of small worlds and implement text clustering quickly. The paper puts forward an idea that adopts inverted index into this algorithm, which simplifies programming and improves operation efficiency. The experimental results verified that the system could find current hot news topics efficiently and users could locate their interested document collection.

Deformation analysis and prediction is a multidisciplinary important research topic. Considering that fuzzy theory has been widely used in prediction methods and models fields at present,the paper introduces fuzzy time series to deformation analysis and prediction. In this paper,the modeling steps of fuzzy time series are briefly introduced and experimental results show that fuzzy time series can be effectively applied to deformation analysis and prediction and improve the accuracy of prediction.

According to the problem that the traditional search algorithms dont consider the needs of individuals, various recommender systems employing different data representations and recommendation methods are currently used to cope with these challenges. In this paper, inspired by the network-based user-item rating matrix, we introduce an improved algorithm which combines the similarity of items with a dynamic resource allocation process. To demonstrate its accuracy and usefulness, this paper compares the proposed algorithm with collaborative filtering algorithm using data from MovieLens. The evaluation shows that, the improved recommendation algorithm based on graph model achieves more accurate predictions and more reasonable recommendation than collaborative filtering algorithm or the basic graph model algorithm does.

There are some problems in the traditional face recognition: too sensitive to light, its classification is too simple, hard to apply to distributed system. For these problems, this paper does three work: used HOG feature in the face recognition, that improved it get more steady feature in different type of light; putting forward NSMD (Nearest Sample Max Distance), it try to find some partly best classifications to unite one strong classification, so the classification hyperplane can be more far from the sample boundary; finally the algorithms of this paper specially the one-to-one design mode are adapt to MapReduce, so its feasible to own the advantages of distributed system (expansively safety...). With the experiment analysis, the work gets their goals: HOG feature can be adapt to different lights, NSMD can get better rate of identification, the MapReduce algorithms can get fine speed-up ratio.

Adaptive learning system has been developed maturing. But the formulation and selection of learning scheme in the learners' personality characteristic mining is still not satisfactory. The adaptive learning scheme with specific learner characteristics personalized will have positive effect for the learner on learning effect. The emergence of adaptive learning system based on network, making the learning is no longer affected by time and space restrictions, and the teachers teaching mode is changing to students' learning mode. The prior paper had given a brief introduction for adaptive learning system. Then in this paper we will focus on the designed theoretical knowledge of the adaptive learning system and give the design scheme.

By investigating insufficiency of typical artificial intelligence algorithms aiming at the high rate of False-Positives and False-Negatives in the Intrusion Detection Systems (IDS), this paper presents an approach that Support Vector Machine (SVM) is embedded in Network Intrusion Detection System (NIDS). At the same time, by using online data and K-fold cross-validation method, this paper proposes a method to optimize the attributes and model of SVM respectively. Experimental results show that by using this method as the detection core of the intrusion detection system, the rate of False-Negatives in IDS can be reduced significantly.

For there is a lower level of fusion grid decision-making process and decision-making are not smart enough ,this paper presents a power grid's management and control task demand – knowledge matching model, to help staff accurately and rapidly access to the necessary knowledge, before making a decision.At the same time,in order to solve the problem ,this paper, in the TBM framework ,presents a grid knowledge clustering-ensemble method. Then, knowledge clustering-ensemble model is applied to the electrical network management control decision-making process. Test of the fault diagnosis has proved that ,the model is able to maintain the original good performance of the evidence theory and TBM framework, also can get more useful information for decision-making.

This paper proposed an improved Web text classification algorithm. It combined SVM algorithm and KNN algorithm and used the KNN algorithm to compensate for the deficiencies of traditional SVM algorithm and took simple ideas and smaller cost to improve traditional SVM algorithm. The experiments show that the proposed algorithm in this paper has gained good effect.

Academic search engines, such as Google Scholar and Scirus, provide a Web-based interface to effectively find relevant scientific articles to researchers. However, current academic search engines are lacking the ability to cluster the search results into a hierarchical tree structure. In this paper, we develop a post-search academic search engine by using a mixed clustering method. In this method, we first adopt a suffix tree clustering and a two-way hash mechanism to generate all meaningful labels. We then develop a divisive hierarchical clustering algorithm to organize the labels into a hierarchical tree. According to the results of experiments, we conclude that using our mixed clustering method to cluster the search results can give significant performance gains than current academic search engines. In this paper, we make two contributions. First, we present a high performance academic search engine based on our mixed clustering method. Second, we develop a divisive hierarchical clustering algorithm to organize all returned search results into a hierarchical tree structure.

Based on the research of extreme learning machine and support vector machine, this paper does the research on the optimization method on extreme learning machine. This paper suggests an optimization model of extreme learning machine based on the improvement of the old model, and this model has obvious improvement on generalization ability and learning parameter ability.

Nowadays, all kinds of technology advance swiftly and vigorously, especially computer technology. In such cases, the computerized test system emerges as the times demand. It will take the place of the traditional test by making use of the computer to manage and possess the item bank, organize the contents and evaluate the test paper. It is characterized in its convenience and fastness in setting a test paper, its speed and accuracy in evaluating the paper and its scientific analysis of the test paper. It is easier for teachers to set a test paper by the item bank system, and it will save a lot of labors in evaluating the test paper. It will greatly improve the accuracy in evaluation. The statistical analysis will be performed automatically by the computer, which will be convenient for students to inquiry about, print, count and analyze their marks. Then in analogy to manual assessment of the exam scripts, an algorithm, by using fuzzy theory to assess the subjective questions, is proposed and analyzed in detail. It is a useful attempt to try to solve the automatic subjective question assessment which would have to be done traditionally by manual effort.

Having already carried on on-line examination under the network environment several years of history, regardless which kind examines it is ,the problems is divided into objective and subjective. It have been very mature to Automatically grade examination papers in the objective by database technique For automatic checking over subjective examination, its procedure involves some restriction theory and technique development,such as artificial intelligence, natural language comprehension and mode identification etc., So far, there is no accepted system.This paper put forward to the research direction of the algorithm in automatic checking over subjective examination by analyzing the present condition for grading examination papers.

In this paper, we consider the problem of evolutionary link strength mining for information diffusion modeling and predicting. Most exiting approaches analyzing link strength focus on either topological information or content semantics, but do not simultaneously consider combining both features. Furthermore, in many applications of text stream data, the users are always in need of the ability to track link strength evolution pattern along time line. In this paper, we argue that the evolutionary link strength reflects a user’s interest to create reply links and is influenced by both neighbors and topic content. With this aim in mind, we propose a novel Content-Topology-Time model for link strength predicting in time-aware online social network. Comprehensive experimental studies on real world micro-blog data set show that our approach outperforms existing ones and well matches the practice.

Category-based statistic language model is an important method to solve the problem of sparse data. But there are two bottlenecks about this model: (1) the problem of word clustering, it is hard to find a suitable clustering method that has good performance and not large amount of computation. (2) class based method always lose some prediction ability to adapt the text of different domain. The authors try to solve above problems in this paper. This paper presents a novel definition of word similarity. Based on word similarity, this paper gives the definition of word set similarity. Experiments show that word clustering algorithm based on similarity is better than conventional greedy clustering method in speed and performance. At the same time, this paper presents a new method to create the vari-gram model.

KNN algorithm is used for Chinese text classification in this paper. First, TF-IDF is chosen as the feature weighting method. To the characteristics of corpus used in this paper, TF-IDF is adjusted to a new method. At last, experimental result shows the accuracy of KNN text classifier can be improved with the adjusted feature weighting method.

This paper presents the empirical results from experiment on speed-accuracy trade-off in ‘drag and drop’ movements. Although the previous work has modeled a theorem, framework and achievement, but some researcher focused only on the system error. In this paper, we design an analysis mechanism with the software that could behave as actual computer in producing error (system bug). This study aims at investigating the user anticipation in multiple modalities in relation to users’ performance. The multiple modalities represent four feedbacks; default, audio, text and visual that integrate with error enforcement forcing error with certain percentage to simulate system error and actual computer behavior. The finding shows the average means of 4.0667 for feedback modality of which, we conclude that the users anticipate with feedback modality can reacts to improve their performance in movement mechanism.

Mulit-agent system[MAS] research on learning has been in the area of negotiation, and learning strategies of other agents.This paper presents an agent learning approach in multi-agent system based on Bayesian learning, it researches to develop agents that learn free-text queries and keyword searches in MAS. The MAS learns to identify an appropriate agent to answer free-text and natural language queries as well as keyword searches submitted by users. The paper describes how Bayesian learning is implemented in MAS, and analyzes the effectiveness of MAS learning based on the Bayesian learning approach by analyzing the accuracy and degree of learning.

The current challenges for enterprises are heavily linked to resource and energy efficiency, a result of political guidelines and the overall consciousness for environmental issues in society. Thus, the impact of energy efficiency on factory planning and production planning is of undeniable importance. Adequate consumption information utilising various depiction methods has become a necessity in this respect. Yet, only few works provide information on the efficiency of energy data acquisition methods. This paper proposes a methodological framework for defining an efficient procedure for energy data acquisition in accordance with the depiction or prognosis method. A preliminary literature review indicates the variety of methods for energy data depiction and prognosis. The classification of these methods regarding planning state and planning levels leads to the estimation of relevant input data. Based on the fundamental investigation of necessary input data the decision approach is developed with a general heuristic decision model and incorporates the Analytic Hierarchy Process for quantification and solution of the overall decision problem. Therefore the integration of multiple decision criteria enables the consideration of different quantitative and qualitative influences. As a result the whole approach supports the collaborative identification of means for energy data acquisition and is applicable in several circumstances where energy data needs to be gathered efficiently. The decision procedures exemplary application has shown the general proficiency of the approach and that excessive data acquisition which contradicts efficiency is avoidable.

Due to the rising energy costs and the increasing competition on the market the improvement of the energy efficiency in production systems can be a great chance for companies to gain an advantage compared to their competitors. Therefore, the transparency of the energy consumption of the several processes of such systems has to be known in detail. In this paper a method for the detection of operating states in production systems, based on big data analysis is introduced. The developed algorithm automatically detects different operating states in the current profile and improves his accuracy the longer it is used. To evaluate the developed data analysis algorithm, a defined process with a conveyor belt was considered and measured several times. The algorithm defines clusters from the measurements and identifies classes for the several operating states. With the naive Bayes classifier it is possible to categorize the clusters more fine-grained and faster than with a correlation function. An advantage of the method is that each new classified instance is taken into account for future unclassified instances. So it is possible to integrate a continuous learning process in production processes and to consider a slowly drift of states. Also, the subsequent addition of classes and attributes that are represented in this work by clusters, is possible at any time.

A wide-spread approach to foster production and logistics related improvements on a system level is value stream mapping (VSM). It is often used as a starting point for implementing lean manufacturing in producing companies. VSM as a modeling approach contains two major phases – one is the analysis, another one the design phase. During these there is always the question of getting accurate information out of the production system to enable a constructive improvement. It can be distinguished between direct data acquisitions (observation, questioning, interviews) and indirect data acquisition (using IT systems) where both quantitative and qualitative data arises. Standard value stream design (VSD) offers no option to combine these data types, instead mostly deterministic values are used for assessing future states; this can lead to an illusion of accuracy. Thus in this paper it is shown how to combine different data types with an appropriate modeling using fuzzy variables in the course of VSD. The whole procedure is applied during a laboratory study to provide internal validity.

Modern data analysis techniques has been successfully applied in many technical disciplines to understand the complexity of the system. The growing volume of theoretical knowledge about systems dynamic's offered researchers the opportunity to look for non-linear dynamics in data whose evolution linear models are unable to explain in a satisfactory manner. One of the most recent approach in this respect is Recurrence Analysis - RA. It is a graphical method designed to locate hidden recurring patterns, nonstationarity and structural changes. RA approach have first been applied in natural sciences like physics and biology. Quickly, was adopted in economics and engineering. Meanwhile, the fast development of computer resources has provided powerful tools to perform this new and complex model. One free software which we used to perform the analysis is Visual Recurrence Analysis - VRA developed by Eugene Kononov. As is presented in this paper, the recurrence plot investigation for the analyzing of the gasoline engine shows some of the its capabilities in this domain. We chose two specific engine parameters measured in two different tests to perform the RPA. These parameters are ignition angle and engine angular speed. There were computed graphs for each of them. Graphs were analyzed and compared to obtain a conclusion. This work is an incipient research, being one of the first attempt of using recurrence plot for analyzing automotive dynamics. It opens a wide field of action for future research programs.

Energy efficiency of production systems and of the product itself has grown to a critical competitive factor. Besides the manufacturer’s monetary motivation there are increasing incentives to meet customers’ expectations regarding lifecycle cost and the ecological footprint of products. That neo-ecology, as one megatrend, leads to a new business moral resulting in an energy optimization of the whole product life cycle in terms of resource and energy input. There is a plenty of measures to reduce the energy consumption of a production system and thus to increase its efficiency. To do so companies do not have to develop proprietary solutions for their production sites but can draw on a large pool of measures. However, in practice, many energy optimization measures are unknown to their energy managers. This is mainly owing to the fact that there is no standardized categorization for energy optimization potentials yet. In addition, many efficiency deficits remain undetected as a result of a non-existing efficient methodology for finding energy consumption optimization measures. The domain of information retrieval addresses this issue, as it is able to provide documents matching the user’s information demand. Nevertheless, search queries have to be sufficiently well known in order to gain adequate results. In this paper we show how ontologies can be used to support the user in defining search queries and finding optimization measures efficiently. As formal and explicit specifications of shared conceptualizations, ontologies offer the possibility to represent relevant parts of knowledge in a standardized, machine-readable manner. Therefore, ontologies improve upon data models, which are mainly used for single applications. For the purpose of energy efficiency in production environments, we provide both a methodology to build ontologies for describing energy saving measures and illustrate the application for explicit energy efficiency optimization measures.

Although, Back Propagation Neural Network are frequently implemented to forecast short-term electricity load, however, this training algorithm is criticized for its slow and improper convergence and poor generalization. There is a great need to explore the techniques that can overcome the above mentioned limitations to improve the forecast accuracy. In this paper, an improved BP neural network training algorithm is proposed that hybridizes simulated annealing and genetic algorithm (SA-GA). This hybrid approach leads to the integration of powerful local search capability of simulated annealing and near accurate global search performance of genetic algorithm. The proposed technique has shown better results in terms of load forecast accuracy and faster convergence. ISO New England data for the period of five years is employed to develop a case study that validates the efficacy of the proposed technique.

Taking the effect of geometric errors on machining accuracy into consideration in the design of small precision machine tools, the thermal errors become the main factor affecting the machining accuracy of machine tools. Currently, how to quickly and effectively evaluate the impact of thermal error on machining precision, in order to make effective thermal error prediction and compensation method, is needed to solve for small precision machining. Based on stepwise regression algorithm theory, combined with the DBB measured and analyzed on MATLAB in this paper, a better thermal error fitting regression formula is obtained. Meanwhile, this mathematical model approach is also applicable in a variety of other areas, such as machine tool thermal error analysis, compensation and testing etc.

In this paper, we present a hierarchical method of activity recognition for sleeping at the desk in business hall. The method consists of three steps. First, the reference points such as body joints are obtained from workers in business hall. Second, we build the dependency graph to represent the relationships between reference points. Third, the multidimensional output regressions along the dependency paths are used to estimate the positions of these reference body points. Experimental results demonstrate that our method achieves comparable accuracy to state-of-the-art results.

In this paper, a design method of applying rational expertise to the learning mechanism based on fuzzy genetic algorithm is proposed. Firstly, we introduce the accuracy and explanatory of the fuzzy classification system, and then the application of expertise to the fuzzy system, and the adjustment of the member function using genetic algorithm to obtain the final classification results. Finally, the classification results of our experiments demonstrate the effectiveness of the design method with the Irvine classification sample data, and we can obtain better accuracy and explanatory through this learning mechanism with the designed fuzzy classification system.

Face recognition has recently become a hot research topic. In order to do more in that area, several algorithms have emerged. However, even the most efficient algorithm has limitations. To overcome this problem, the combination of algorithms is sometimes used. In this paper a methodology based on two approaches is presented. Firstly, we use Locality Preserving Projection (LPP) for feature extraction. Secondly, the Back Propagation Neural Network (BPNN) is used for recognition. Experiments have been done using 400 images of ORL database. Experimental results show that the algorithm is performs well and achieves good recognition.

There are some bottleneck problems in the supervised machine learning and unsupervised machine learning. In view of the current problems, this paper tries to make some meaningful exploration. The main work is as follows: Research on the statistical analysis of factor analysis and latent variable and in some valuable research results of typical machine learning, and some no analysis method and factor analysis of supervised learning or hidden variables method to contact with the typical analysis, summary of the comprehensive characteristics of implicit factor model and to reveal the hiding data structures help and contributions.

This paper studied the application of fuzzy comprehensive evaluation in meteorological mutation identification of load forecasting. It measured out a specific algorithm in meteorological mutation identification of ultra-short-term load forecasting, so that to improve the load forecasting accuracy and reduce the relative error. When using the fuzzy comprehensive evaluation method, set the membership function and factor right based on the characteristics of a province for the summer weather factors, and measure the mutation membership value with the appropriate operator, which can provide a reference for the study of meteorology identification. The result shows that using fuzzy comprehensive evaluation to identify meteorological mutation has good usability and accuracy.

Confidence regression is a significant research field of confidence machine learning. This paper adopts KNN algorithm as a tool, and performs error evaluation on results of regressive learning to classify the accept field and the refuse field so as to achieve the confidence regression. By setting specific error value, this approach achieves controllable confidence regression, which has been tested on experimental data of bodyfat and other data sets. The experimental results presented show the feasibility of our approach.

Network technology accelerates the development of educational information, campus portal building is considered as an important part of it in every university, almost all information of teaching and research appeared on the web. Meanwhile, the utilization rate of some websites was lower in university, information was updated slowly, information classifications were complex and not standardized on a platform. They didn’t emphasis on using and sharing but building and developing, and this phenomenon was widespread. So the paper proposed a decision tree model for score sorting information based on C5.0 algorithm, setting up a statistical model for data mining by adding a line weight value for portal information. Finally, the results verify the correctness and science of the model by giving an example.

This paper presents the selection methods of typical substations based on grey correlation analysis method, and proposes classification principles of load characteristics. The actual example shows that using this method in process of statistical synthesis method modelling, the typical substations selected could be more representative, and the popularization and application of the modelling results are more accurate and effective.

As one of the most popular and effective classification algorithms, Support Vector Machine (SVM) has attracted much attention in recent years. Classifiers ensemble is a research direction in machine learning and statistics, it often gives a higher classification accuracy than the single classifier. This paper proposes a new ensemble algorithm based on SVM. The proposed classification algorithm PB-SVM Ensemble consists of some SVM classifiers produced by PCAenSVM and fifty classifiers trained using Bagging, the results are combined to make the final decision on testing set using majority voting. The performance of PB-SVM Ensemble are evaluated on six datasets which are from UCI repository, Statlog or the famous research. The results of the experiment are compared with LibSVM, PCAenSVM and Bagging. PB-SVM Ensemble outperform other three algorithms in classification accuracy, and at the same time keep a higher confidence of accuracy than Bagging.

Frequent patterns mining is one of the most important tasks in data mining, traditional algorithms usually deal with this problem in simple structured data, but there are so much complex data in reality, for example, the tree type of data, graph type of data, and so on, when investigating these complex structured data, constrains are often needed to be given in order to narrow the search space, however, this will lose some of the useful interesting patterns. In this paper, we present a novel algorithm based on higher-order logic to discover frequent patterns in complex structured data, the novel method can overcome some drawbacks occurring in traditional algorithms. We use Escher, which is a higher-order logic programming language, to discover frequent patterns in complex structured data. Experimental results show that the proposal algorithm is efficient and scalable.

The real-time prediction of network security situation can significantly improve the monitoring and emergency response capability of the network. However, in practice, if there are a large amount of false predictions, the network administrators should become insensitive and will finally ignore all prediction results. In this paper, we try to solve this issue and propose a novel False Positive Adaptive (FPA) method for network security situation prediction. The main idea of our method is using extrainformation to reduce the number of false positives in prediction. In the model training step, we take advantage of host and network information to eliminate meaningless alerts produced by security tools such as Intrusion Detection System (IDS) and firewall, thus assuring the accuracy of the training samples. In the prediction step, we utilize the detection information from security tools to confirm the prediction results automatically. If the previous predictions are not detected, they will be considered as false positives and the prediction model will be retrained by incremental learning. In our work, the model training and incremental learning is accomplished efficiently by neural network and boosting algorithm.

Due to the problem that the existing topic detection algorithms can not satisfy accuracy,real time and topic hierarchical clustering at the same time, this article builds a hierarchy topic detection algorithm based on improved single pass clustering algorithm. In addition, using public opinion evaluation indexes to analyze topic temperature,the method proposed in this paper can detect hot topics accurately and timely while showing the hierarchical structure of the topic .

We propose a new generative probabilistic Dirich- let Author-Topic (DAT) Model for extracting information about authors and topics from large text collections. DAT is a three-level hierarchical Bayesian model. The model builds on the Author Topic (AT) model, adding the key attribute that distribution over author is conditioned on a Dirichlet prior. The probability distribution over topics in a multi-author document is a mixture of the distributions associated with the authors. The three level distributions including document-author, author-topic and topic-word are learned from data in an unsupervised manner using a Gibbs sampling algorithm. We give results on a large corpus which contains 1740 papers from the Neural Information Processing Systems Conference (NIPS). Experiments based on perplexity scores for test documents are used to illustrate systematic differences between the proposed model and a number of alternatives.