The concept of `diversity' has been one of the main open issues in the field of multiple classifier systems. In this paper we address a facet of diversity related to its effectiveness for ensemble construction, namely, explicitly using diversity measures for ensemble construction techniques based on the kind of overproduce and choose strategy known as ensemble pruning. Such a strategy consists of selecting the (hopefully) more accurate subset of classifiers out of an original, larger ensemble. Whereas several existing pruning methods use some combination of individual classifiers' accuracy and diversity, it is still unclear whether such an evaluation function is better than the bare estimate of ensemble accuracy. We empirically investigate this issue by comparing two evaluation functions in the context of ensemble pruning: the estimate of ensemble accuracy, and its linear combination with several well-known diversity measures. This can also be viewed as using diversity as a regularizer, as suggested by some authors. To this aim we use a pruning method based on forward selection, since it allows a direct comparison between different evaluation functions. Experiments on thirty-seven benchmark data sets, four diversity measures and three base classifiers provide evidence that using diversity measures for ensemble pruning can be advantageous over using only ensemble accuracy, and that diversity measures can act as regularizers in this context.

The paper considers the problem of increasing the generalization ability of classification systems by creating an ensemble of classifiers based on the CNN architecture. Different structures of the ensemble will be considered and compared. Deep learning fulfills an important role in the developed system. The numerical descriptors created in the last locally connected convolution layer of CNN flattened to the form of a vector, are subjected to a few different selection mechanisms. Each of them chooses the independent set of features, selected according to the applied assessment techniques. Their results are combined with three classifiers: softmax, support vector machine, and random forest of the decision tree. All of them do simultaneously the same classification task. Their results are integrated into the final verdict of the ensemble. Different forms of arrangement of the ensemble are considered and tested on the recognition of facial images. Two different databases are used in experiments. One was composed of 68 classes of greyscale images and the second of 276 classes of color images. The results of experiments have shown high improvement of class recognition resulting from the application of the properly designed ensemble.

Complex circuit of milling-classify systems are used in different branches of industry, because the required particle size distribution of product can seldom be reached in a single-stage grinding on the same device. The multistage processes of comminution and classification make possible suitable selection of parameters process for variables graining of fed material, mainly through sectioning of devices or change of their size and the types. Grinding material usually contains size fractions, which meet the requirements relating finished product. Then profitable is preliminary distributing material on a few size fractions, so to deal out with them demanded fraction of product, whereas remaining to direct alone or together with fed material to the same or different device. If the number of mills and classifiers in a circuit is large enough, building the model of particle size distribution transformation becomes rather complicated even for the circuit of a given structure. The situation becomes much more complicated, if we want to compare characteristics of all possible circuits, that can be constructed from these mills and classifiers, because the number of possible circuits increases greatly with the increase of number of devices being in the milling-classify system. The method creating matrix model for transformation of particle size distribution in a circuit of arbitrary structure of milling-classify system is presented in the article. The proposed model contains the mass population balance of particle equation, in which are block matrices: the matrix of circuit M, the matrix of inputs F and the matrix of feed F0. The matrix M contains blocks with the transition matrix P, the classification matrix C, the identity matrix I and the zero matrix 0 or elements describing the transformation of particle size distribution in the circuit. The matrix F is the block column matrix, which elements describing all particle size distributions at inputs to the circuit elements. The matrix F0 is the block column matrix, which elements describing particle size distributions in all feeds to the circuit. In paper was discussed this model in details, showed algorithm and three examples formatrix construction for the closed circuit ofmilling-classify systems. In conclusion was affirmed, that presented model makes possible to forecasting particle size distribution of grinding product, which leaving chosen the unit of system. The matrix model can be applied to improving modeling of mineral processing in the different grinding devices.

When the distribution of water quality samples is roughly balanced, the Bayesian criterion model of water-inrush source generally can obtain relatively accurate results of water-inrush source identification. However, it is often difficult to achieve desired classification results when training samples are imbalanced. Sample imbalance is common in the source identification of mine water-inrush. Therefore, we propose a three-dimensional (3D) spatial resampling method based on rare water quality samples, which achieves the balance of water quality samples. Based on the virtual water sample points distributed by the 3D grid, the method uses the 3D Inverse Distance Weighting (IDW) method to interpolate the groundwater ion concentration of the virtual water samples to achieve oversampling of rare water samples. Case study in Gubei Coal Mine shows that the method improves overall discriminant accuracy of the Bayesian criterion model by 5.26%, from 85.26% to 90.69%. In particular, the discriminative precision of the rare class is improved from 0% to 83.33%, which indicates that the method can improve the discriminant accuracy of the rare class to large extent. In addition, this method increases the Kappa coefficient of the model by 19.92%, from 52.26% to 72.19%, increasing the degree of consistency from “general” to “significant”. Our research is of significance to enriching and improving the theory of prevention and treatment of mine water damage.

Radial basis function neural networks (RBF NNs) are one of the most useful tools in the classification of the sonar targets. Despite many abilities of RBF NNs, low accuracy in classification, entrapment in local minima, and slow convergence rate are disadvantages of these networks. In order to overcome these issues, the sine-cosine algorithm (SCA) has been used to train RBF NNs in this work. To evaluate the designed classifier, two benchmark underwater sonar classification problems were used. Also, an experimental underwater target classification was developed to practically evaluate the merits of the RBFbased classifier in dealing with high-dimensional real world problems. In order to have a comprehensive evaluation, the classifier is compared with the gradient descent (GD), gravitational search algorithm (GSA), genetic algorithm (GA), and Kalman filter (KF) algorithms in terms of entrapment in local minima, the accuracy of the classification, and the convergence rate. The results show that the proposed classifier provides a better performance than other compared classifiers as it classifies the sonar datasets 2.72% better than the best benchmark classifier, on average.

The paper presents special forms of an ensemble of classifiers for analysis of medical images based on application of deep learning. The study analyzes different structures of convolutional neural networks applied in the recognition of two types of medical images: dermoscopic images for melanoma and mammograms for breast cancer. Two approaches to ensemble creation are proposed. In the first approach, the images are processed by a convolutional neural network and the flattened vector of image descriptors is subjected to feature selection by applying different selection methods. As a result, different sets of a limited number of diagnostic features are generated. In the next stage, these sets of features represent input attributes for the classical classifiers: support vector machine, a random forest of decision trees, and softmax. By combining different selection methods with these classifiers an ensemble classification system is created and integrated by majority voting. In the second approach, different structures of convolutional neural networks are directly applied as the members of the ensemble. The efficiency of the proposed classification systems is investigated and compared to medical data representing dermoscopic images of melanoma and breast cancer mammogram images. Thanks to fusion of the results of many classifiers forming an ensemble, accuracy and all other quality measures have been significantly increased for both types of medical images.

The binary classifiers are appropriate for classification problems with two class labels. For multi-class problems, decomposition techniques, like one-vs-one strategy, are used because they allow the use of binary classifiers. The ensemble selection, on the other hand, is one of the most studied topics in multiple classifier systems because a selected subset of base classifiers may perform better than the whole set of base classifiers. Thus, we propose a novel concept of the dynamic ensemble selection based on values of the score function used in the one-vs-one decomposition scheme. The proposed algorithm has been verified on a real dataset regarding the classification of cutting tools. The proposed approach is compared with the static ensemble selection method based on the integration of base classifiers in geometric space, which also uses the one-vs-one decomposition scheme. In addition, other base classification algorithms are used to compare results in the conducted experiments. The obtained results demonstrate the effectiveness of our approach.

Construction planning always requires labour productivity estimation. Often, in the case of monolithic construction works, the available catalogues of productivity rates do not provide a reliable assessment. The paper deals with the problem of labour estimation for reinforcement works. An appropriate model of labour prediction problem is being introduced. It includes, between others, staff experience and reinforcement buildability. In the paper it is proposed, that labour requirements can be estimated with aggregated classifiers. The work is a continuation of earlier studies, in which the possibility of using classifier ensembles to predict productivity in monolithic works was investigated.

Customer churn prediction is used to retain customers at the highest risk of churn by proactively engaging with them. Many machine learning-based data mining approaches have been previously used to predict client churn. Although, single model classifiers increase the scattering of prediction with a low model performance which degrades reliability of the model. Hence, Bag of learners based Classification is used in which learners with high performance are selected to estimate wrongly and correctly classified instances thereby increasing the robustness of model performance. Furthermore, loss of interpretability in the model during prediction leads to insufficient prediction accuracy. Hence, an Associative classifier with Apriori Algorithm is introduced as a booster that integrates classification and association rule mining to build a strong classification model in which frequent items are obtained using Apriori Algorithm. Also, accurate prediction is provided by testing wrongly classified instances from the bagging phase using generated rules in an associative classifier. The proposed models are then simulated in Python platform and the results achieved high accuracy, ROC score, precision, specificity, F-measure, and recall.