Standard UCI Datasets Research Articles

Modified method for wavelet neural network model

To improve the performance of Wavelet Neural Network( WNN) model in dealing with complex nonlinear problems, concerning the shortcomings of premature convergence, poor late diversity, poor search accuracy of Quantumbehaved Particle Swarm Optimization( QPSO) algorithm, a modified quantum-behaved particle swarm algorithm was proposed for WNN training by introducing weighting coefficients, introducing Cauchy random number, improving contraction-expansion coefficient and introducing natural selection at the same time. And then, it replaced the gradient descent method with the modified quantum-behaved particle swarm algorithm, trained the wavelet coefficients and network weights, and then input the optimized combination of parameters into wavelet neural network to achieve the algorithm coupling. The simulation results on three UCI standard datasets show that the running time of the Modified Quantum-behaved Particle Swarm Optimization-Wavelet Neural Network( MQPSO-WNN) was reduced by 11% ~ 43%, while the calculation error was decreased by 8% ~ 57%,compared with wavelet neural network, PSO-WNN and QPSO-WNN. Therefore, the MQPSO-WNN model can approximate the optimal value more quickly and more accurately.

Multi-view semi-supervised collaboration classification algorithm with combination of agreement and disagreement label rules

To improve the performance of the co-training algorithm and expand the range of applications, a multi-view semi-supervised collaboration classification algorithm with the combination of consistent and inconsistent label rules was proposed, which aimed at providing a more effective method for the classification of the bridge structured health data. The proposed algorithm used combination of agreement and disagreement label rules for the unlabeled data by judging whether the two classifiers were consistent. Put the sample to the label set, if the label results were consistent. If the label results were inconsistent and the confidence was beyond the threshold, it put the label result of the high confidence to the label set, took full use of the unlabeled data to improve the performance of the classifier, and updated the classification model by the difference of the classifiers. The experimental results of the proposed algorithm on the bridge structured health datasets and standard UCI datasets verify the effectiveness and feasibility of the proposed model on the multi-view classification problems.

A hybrid genetic algorithm for feature subset selection in rough set theory

Rough set theory has been proven to be an effective tool to feature subset selection. Current research usually employ hill-climbing as search strategy to select feature subset. However, they are inadequate to find the optimal feature subset since no heuristic can guarantee optimality. Due to this, many researchers study stochastic methods. Since previous works of combination of genetic algorithm and rough set theory do not show competitive performance compared with some other stochastic methods, we propose a hybrid genetic algorithm for feature subset selection in this paper, called HGARSTAR. Different from previous works, HGARSTAR embeds a novel local search operation based on rough set theory to fine-tune the search. This aims to enhance GA's intensification ability. Moreover, all candidates (i.e. feature subsets) generated in evolutionary process are enforced to include core features to accelerate convergence. To verify the proposed algorithm, experiments are performed on some standard UCI datasets. Experimental results demonstrate the efficiency of our algorithm.

Finding Rare Classes: Active Learning with Generative and Discriminative Models

Discovering rare categories and classifying new instances of them are important data mining issues in many fields, but fully supervised learning of a rare class classifier is prohibitively costly in labeling effort. There has therefore been increasing interest both in active discovery: to identify new classes quickly, and active learning: to train classifiers with minimal supervision. These goals occur together in practice and are intrinsically related because examples of each class are required to train a classifier. Nevertheless, very few studies have tried to optimise them together, meaning that data mining for rare classes in new domains makes inefficient use of human supervision. Developing active learning algorithms to optimise both rare class discovery and classification simultaneously is challenging because discovery and classification have conflicting requirements in query criteria. In this paper, we address these issues with two contributions: a unified active learning model to jointly discover new categories and learn to classify them by adapting query criteria online; and a classifier combination algorithm that switches generative and discriminative classifiers as learning progresses. Extensive evaluation on a batch of standard UCI and vision data sets demonstrates the superiority of this approach over existing methods.

Active AODE learning based on a novel sampling strategy and its application

In this paper, we first present active Averaged One-Dependence Estimator AODE learning classification model, which can improve the performance of AODE by selecting and asking experts to label the samples only with maximum information. Several common sampling strategies for active learning are discussed. Unfortunately, these methods can get the outlier, which will lead to scale up the classification-reduced error and high complexity. Motivated by those analyses, we propose a new active learning strategy, which is based on the uncertainty sampling and classification accuracy loss sampling strategy. Experimental results on three UCI standard data sets and a real remote sensing data set show that the AODE classification model and our novel active learning strategy can get better classification accuracy with fewer labelled samples than that of the state-of-the-art approaches for active learning.

A Method of Multi-Classifier Combination Based on Dempster-Shafer Evidence Theory and the Application in the Fault Diagnosis

A problem is aroused in multi-classifier system that normally each of the classifiers is considered equally important in evidences’ combination, which gone against with the knowledge that different classifier has various performance due to diversity of classifiers. Therefore, how to determine the weights of individual classifier in order to get more accurate results becomes a question need to be solved. An optimal weight learning method is presented in this paper. First, the training samples are respectively input into the multi-classifier system based on Dempster-Shafer theory in order to obtain the output vector. Then the error is calculated by means of figuring up the distance between the output vector and class vector of corresponding training sample, and the objective function is defined as mean-square error of all the training samples. The optimal weight vector is obtained by means of minimizing the objective function. Finally, new samples are classified according to the optimal weight vector. The effectiveness of this method is illustrated by the UCI standard data set and electric actuator fault diagnostic experiment.

Leveraging k-NN for generic classification boosting

Voting rules relying on k-nearest neighbors (k-NN) are an effective tool in countless many machine learning techniques. Thanks to its simplicity, k-NN classification is very attractive to practitioners, as it enables very good performances in several practical applications. However, it suffers from various drawbacks, like sensitivity to “noisy” instances and poor generalization properties when dealing with sparse high-dimensional data.In this paper, we tackle the k-NN classification problem at its core by providing a novel k-NN boosting approach. Namely, we propose a supervised learning algorithm, called Universal Nearest Neighbors (UNN), that induces a leveraged k-NN rule by globally minimizing a surrogate risk upper bounding the empirical misclassification rate over training data. Interestingly, this surrogate risk can be arbitrary chosen from a class of Bregman loss functions, including the familiar exponential, logistic and squared losses. Furthermore, we show that UNN allows to efficiently filter a dataset of instances by keeping only a small fraction of data.Experimental results on the synthetic Ripley's dataset show that such a filtering strategy is able to reject “noisy” examples, and yields a classification error close to the optimal Bayes error. Experiments on standard UCI datasets show significant improvements over the current state of the art.

Approaches to Knowledge Reduction of Decision Systems based on Conditional Rough Entropy

Knowledge reduction in rough set theory is an important feature selection method. Since it is an NP-hard problem, it is necessary to investigate fast and effective approximate algorithms. In this paper, to address this issue, by introducing rough entropy in information systems, the novel measures of conditional rough entropy with distinguishing consistent objects form inconsistent objects are presented for both consistent and inconsistent decision systems. Thus, many important propositions, properties, and conclusions for reduct are drawn, and by using decomposition, radix sorting, hash, and input sequence techniques, we construct a forward greedy algorithm for knowledge reduction. Finally, through analyzing the given example, compared with some standard UCI datasets and other knowledge reduction algorithms, the proposed technique is effective and suitable for both consistent and inconsistent decision systems. Thus, it establishes the theoretical basis for seeking efficient algorithm of knowledge acquisition in decision systems.

A rough set approach to feature selection based on ant colony optimization

Rough set theory is one of the effective methods to feature selection, which can preserve the meaning of the features. The essence of rough set approach to feature selection is to find a subset of the original features. Since finding a minimal subset of the features is a NP-hard problem, it is necessary to investigate effective and efficient heuristic algorithms. Ant colony optimization (ACO) has been successfully applied to many difficult combinatorial problems like quadratic assignment, traveling salesman, scheduling, etc. It is particularly attractive for feature selection since there is no heuristic information that can guide search to the optimal minimal subset every time. However, ants can discover the best feature combinations as they traverse the graph. In this paper, we propose a new rough set approach to feature selection based on ACO, which adopts mutual information based feature significance as heuristic information. A novel feature selection algorithm is also given. Jensen and Shen proposed a ACO-based feature selection approach which starts from a random feature. Our approach starts from the feature core, which changes the complete graph to a smaller one. To verify the efficiency of our algorithm, experiments are carried out on some standard UCI datasets. The results demonstrate that our algorithm can provide efficient solution to find a minimal subset of the features.