Search

Classification problem is one of the essential tasks in data mining. Traditional classification strategies are predominantly via cost-insensitive equilibrium data. They tend to be concentrated on the overall accuracy of a model, and such classifiers are improper for unbalanced sample data. Hence, optimizing unbalanced samples to improve classifier performance is an issue worthy of discussion. Based on the information-rich minority samples that are difficult to learn, Majority Weighted Minority Oversampling Technique (MWMOTE) uses the clustering method to generate synthetic samples from the weighted information samples. However, the accuracy of the clustering should be optimized. To this end, a method called NC_Link_MWMOTE is presented for efficiently handling imbalanced learning problems. We propose a solution by using NC_Link-based hierarchical clustering method to synthesize different samples from a small number of samples, thus optimizing the clustering effect. NC_Link_MWMOTE was evaluated on six different levels of equilibrium data sets. The simulation results show that our method is effective and outperforms competitive baseline method in terms of various assessment metrics, such as Fl-score and Area Under Curve (AUC).

This paper proposes a novel method based on the language representation model called BERT (Bidirectional Encoder Representations from Transformers) for Obstetric assistant diagnosis on Chinese obstetric EMRs (Electronic Medical Records). To aggregate more information for final output, an enhanced layer is augmented to the BERT model. In particular, the enhanced layer in this paper is constructed based on strategy 1(A strategy) and/or strategy 2(A-AP strategy). The proposed method is evaluated on two datasets including Chinese Obstetric EMRs dataset and Arxiv Academic Paper Dataset (AAPD). The experimental results show that the proposed method based on BERT improves the F1 value by 19.58% and 2.71% over the state-of-the-art methods, and the proposed method based on BERT and the enhanced layer by strategy 2 improves the F1 value by 0.7% and 0.3% (strategy 1 improves the F1 value by 0.68% and 0.1%) over the method without adding enhanced layer respectively on Obstetric EMRs dataset and AAPD dataset.

Labeling test cases is expensive. Given a limited number of test cases to be labeled, more evenly spreading test cases over the input domain has a better chance to hit the nonpoint failure patterns. In this paper, we propose a spreading points (test cases) algorithm based on local layout of points (the locally spreading approach is referred to as LS). The LS repositions the initial test set and evolves it to improve the minimum distance among points. During the locally spreading process, for every point, a feasible direction of movement is solved according to its nearest neighbors, and along this direction the point can increase the shortest pairwise distance between it and other points. We investigate the effective of the LS approach, considering it as an add-on to the Adaptive Random Testing (ART) technique. The simulation results show that the LS can improve effectiveness (P-measure) of the ART.

In recent years, clustering has become a hotspot in the field of data mining, as one of the key technologies of getting data distribution and observing the characteristics of class. However, some clustering algorithms depend on the selection of initial clustering centers, and the clustering results easily fall into local optimal. To solve the above problem, the paper integrates differential evolution algorithm and adaptive opposition-based learning. The algorithm makes use of reverse factor to guide algorithm search space approaching to the global optimal solution in each generation. In this paper, the improved algorithm is combined with classical K-means algorithm. According to the result of the three sets of data from UCI data verification, it demonstrates that the improved clustering algorithm can not only cluster better and converge faster, but also effectively suppress the occurrence of prematurity.

This paper presents a method for extracting discriminative key poses for skeleton-based action recognition. Poses are represented by normalized joint locations, velocities and accelerations of skeleton joints. An extended label consistent K-SVD (ELC-KSVD) algorithm is proposed for learning the common and action-specific dictionaries. Discriminative key poses are represented by the atoms of the action-specific dictionaries. With the specific dictionaries, sparse codes are obtained for representing action instances through max pooling and temporal pyramid. A SVM classifier is trained for action recognition. The proposed method was evaluated on the MSRC-12 gesture and MSR-Action 3D datasets. Experimental results have shown that the proposed method is effective in extracting discriminative key poses.

Recently, Zhang and Huang [Convergence of a generalized MSSOR method for augmented systems, J. Comput. Appl. Math. 236 (2012) 1841-1850] discussed the GMSSOR method. In this paper, we further study the generalized modified symmetric successive (GMSSOR) method for solving singular augmented systems, and give the corresponding semi-convergence analysis.

This paper starting from the teacher assessment data, analyze whether the evaluation indicators are suitable for factor analysis, then find the two factors which played a decisive role in evaluation indicators these indicators were interactive and related, so as to achieve the purpose of simplifying calculation. Final get the teachers' rankings by factor analysis.

Wireless sensor networks may be used in the explosion environment, such as underground tunnel in coal mine. In order to assure sensor node work normally, the power of a sensor node must be intrinsic safe. The intrinsic safe power of a sensor node, which consists of power supply module and intrinsic safe protection module, is developed. The power supply mainly consists of current-mode controller chip LTC1871 which assures a stable output voltage of a sensor node. The intrinsic safe protection consists of a dual-stage electric current monitor chip and a voltage regulator tube, which are combined with an over-voltage protection that was integrated in LTC1871 to make up intrinsic safe protection circuit of double over-voltage and over-current protection. Theory analysis and simulation test on discharge process of intrinsic safe power are described. The results display that the discharge energy is nonlinear proportion to the current, increases with the current increasing, and the maximum discharge energy is less than 0.02mJ; the discharge power is linear proportion to the current, increases with the current increasing, and the maximum discharge power is less than 60W, which meets the intrinsic safe standard, therefore the power of the sensor node is intrinsic safe.

This dissertation proposes a new algorithm of distributed mining association rules using the improved Apriori algorithm, based on analyses and introduction of the basic concepts and algorithms of mining association rules and mining association rules in distributed databases. Using improved Apriori algorithm to directly produce all of local frequent itemset in each crunode, rather than iteratively selecting candidate itemset. Then gather all of local multifarious itemset to broadcast to the general node, producing the global frequent itemset of association rules. In the process, the data is no longer saved with the affair ID as the key word. We take the item ID as the new key word. The performance of the improved Apriori algorithm has been improved through cutting down the store space. While the general node gathers all of local frequent itemset to select the global frequent itemset, it needs only a broadcast probably, needing three broadcasts worst. This raised the efficiency of the new algorithm of Association Rules in Distributed Database System.

According to the stored mode of massive data in the relational database, this paper proposed a fast mining algorithm to find maximum frequent item sets based on item sequence set grid space. The traditional methods for mining association rules generate frequent item sets from small to large. These approaches are either time consuming or computationally expensive, and often generate a large number of redundant candidates or frequent item sets, which is fatal for controlling mining speed as data to mass-level. The goal of this paper is first to use a self-defined structure linked list to storage item sequence then to find the frequent item sets from large to small. Several applications of association rules mining using item sequence set grid space has a good performance but it demonstrated inefficiency in massive data mining. The problem involves time spent on sub item sets finding. Experimental results will be presented to show that the fast mining algorithm ISSDL-DM proposed in this paper use much less time than the similar existing algorithm ISS-DM for achieving the same outcomes.

As a simple, effective and nonparametric classification method, k Nearest Neighbor (KNN) is widely used in document classification for dealing with the much more difficult problem such as large-scale or many of categories. But KNN classifier may have a problem when training samples are uneven. The problem is that KNN classifier may decrease the precision of classification because of the uneven density of training data. To solve the problem, a new clustering-based KNN method is presented in this paper. It preprocesses training data by using clustering , then classify with a new KNN algorithm, which adopts a dynamic adjustment in each iteration for the neighborhood number K.This method would avoid the uneven classification phenomenon and reduce the misjudgment of the boundary testing samples. We have an experiment in text classification and the result shows that it has good performance.

A large number of materialized views are stored in data warehouse to enable users to quickly get search results for OLAP analysis. But when the remote basic data source changes, the materialized views in data warehouse are also updated correspondingly in order to maintain the consistency with basic relations, which causes materialized views maintenance issues. There are two methods for materialized views maintenance. One way is to re-compute the views, which can lead to extra large storage and maintenance cost and is sometimes unachievable due to storage limitation. So incremental maintenance technique is more preferable in recent years. Its principle is that data source reports its changes to the integrator who then calculates the corresponding changes and inform the database with the results. Incremental maintenance technique is adopted in this paper. The amount of incremental data is different for the same view when adopting different methods, which result in different maintenance costs. The idea and strategy of minimum incremental maintenance is presented. The materialized view definitions and maintenance expressions, as well as algorithms are given. The experiment shows that the maintenance cost of materialized views is decreased and data warehouse processing efficiency is improved.

With the popularization of information and the establishment of the databases in great number, and how to extract data from the useful information is the urgent problem to be solved. Machine learning is the core issue of artificial intelligence research, this paper introduces the definition of machine learning and its basic structure, and describes a variety of machine learning methods, including rote learning, inductive learning, analogy learning , explained learning, learning based on neural network and knowledge discovery and so on. This paper also brings foreword the objectives of machine learning, and points out the development trend of machine learning. Keywords-machine learning; intelligence; methods; application

In this paper classification of data mining based on radial basis function neural networks is researched. After intensive analysis, the training algorithm of radial basis function neural networks is improved in optimum structure, learning speed and approximation accuracy. In learning speed, two-stage learning strategy is used to accelerate the learning process. In approximation accuracy, an error-correction algorithm is presented to improve the output accuracy of radial basis function. In optimum structure, the paper is focused on the number and center selection of the hidden layer units and proposes an adaptive dynamic and static combination algorithm of center selection. Finally, the algorithms are experimented and comparative analyzed. The experimental results show that the performance of the algorithm is significantly improved, and also prove the validity of the improved algorithm.

With the needs of decision-support information of enterprise and the fast development of computer technologies data warehouse technology come out. The data warehouse is a repository of information collected from multiple, possibly heterogeneous, autonomous, distributed databases. The information stored at the data warehouse is in form of views referred to as materialized views. The design of data warehouse is one of the core research problems in studying and evolution of data warehouse. One of the most important decisions in design of data warehouse is the data warehouse selection. Selecting views to materialize impacts on the efficiency as well as the total cost of establishing and running a data warehouse. So, we develop algorithms to select a set of views to materialize in data warehouse in order to minimize the total view maintenance cost under the constraint of a given query response time. We call it query cost view selection problem (QC_VSP). In this paper, First, we propose query cost view selection problem model. Second, we give three algorithms for QC_VSP; we give view_node_matrix in order to solve it. Third, experiment simulation is adopted. The results show that our algorithm works better in practical cases. We implemented our algorithms and a performance study of the algorithms shows that the proposed algorithm delivers an optimal solution. Finally, we discuss the observed behavior of the algorithms. We also identify some important issues for future investigations.

The data warehouse is a repository of information collected from multiple, possibly heterogeneous, autonomous, distributed databases. The information stored at the data warehouse is in form of views, referred to as materialized views. One of the most important decisions in designing a data warehouse is selection of right views to be materialized. So, we develop algorithms to select a set of views to materialize in data warehouse in order to minimize the total view maintenance cost under the constraint of a given query response time. We call it query cost view_ selection problem (QC_VSP). In this paper, First, we propose the cost model of QC_VSP. Second, we design algorithms for QC_VSP. Third, we use experiments do demonstrate the power of our approach.

In this paper an Inverse Synthetic Aperture Radar (ISAR) imaging algorithm for ship targets based on the division of imaging intervals and the selection of optimum time is proposed. The relative motion of a ship on the ocean wave can be broken into three components, namely pith, roll and yaw, which makes the Doppler frequency vary with the slow time. So during the whole slow time the echoes are not linear frequency modulation (LFM) signals any more. Under this condition the division of the imaging intervals and the selection of the optimum imaging time during the considered imaging interval are of great importance for ISAR imaging of ship targets. The imaging results of the measured data demonstrate the effectiveness of the proposed approach.

Stepped-frequency subpulse signals are widely used to obtain ultra-high range resolution. The stepped-frequency subpulse signals can be combined to one single signal with high bandwidth by using synthetic bandwidth methods. In practical SAR imaging motion error and the time delay of echo signal must be considered before applying the available synthetic bandwidth methods. This paper presents a method in which motion compensation and compensation in Doppler domain are integrated with time domain synthetic bandwidth method in order to get high quality SAR image.

In this paper, we first propose some new and universal features of all kinds of P2P traffic derived from the packet header information of transfer/network layer and present a novel approach for gathering P2P traffic: just from a single host perspective, namely capturing packets from the NIC (network interface card) of one host.

A longitudinal impedance measurement system was established for the BEPCII. The measurements, done in the frequency domain, are based on the coaxial wire method using HP/Agilent 8720ES network analyzer. The applications of the TRL calibration technique and absorbers were investigated to find a good approach for impedance measurements. The impedance, larger than 20 Ohm and below 6 GHz, can be measured using the TRL calibration technique in the experiment. And better measurement results were got using the reference pipes with the absorbers. So, this system satisfies the requirements of the BEPCII. This paper gives a review on this impedance measurements system for the BEPCII. The measurements results show that there are no serious impedance problems for BEPCII bellows and injection kickers, agreeing well with the numerical simulations. More improvements on this system are in progress.