Search

Nowadays, decision support systems (DSSs) are widely used in several application domains, from industrial to healthcare and medicine fields. Concerning the industrial scenario, we propose a DSS oriented to the aesthetic quality control (AQC) task, which has quickly established itself as one of the most crucial challenges of Industry 4.0. Taking into account the increasing amount of data in this domain, the application of machine learning (ML) and deep learning (DL) techniques offers great opportunities to automatize the overall AQC process. State-of-the-art is mainly oriented to approach this problem with a nominal DL classification method which does not exploit the ordinal structure of the AQC task, thus not penalizing the error among distant AQC classes (which is a relevant aspect for the real use case). The paper introduces a DL ordinal methodology for the AQC classification. Differently from other deep ordinal methods, we combined the standard categorical cross-entropy with the cumulative link model and we imposed the ordinal constraint via the thresholds and slope parameters. Experimental results were performed for solving an AQC task on a novel image dataset originated from a specific company’s demand (i.e., aesthetic assessment of wooden stocks). We demonstrated how the proposed methodology is able to reduce misclassification errors (up to 0.937 quadratic weight kappa loss) among distant classes while overcoming other state-of-the-art deep ordinal models and reducing the bias factor related to the item geometry. The proposed DL approach was integrated as the main core of a DSS supported by Internet of Things (IoT) architecture that can support the human operator by reducing up to 90% the time needed for the qualitative analysis carried out manually in this specific domain.

Modelling extreme values distributions, such as wave height time series where the higher waves are much less frequent than the lower ones, has been tackled from the point of view of the Peak-Over-Threshold (POT) methodologies, where modelling is based on those values higher than a threshold. This threshold is usually predefined by the user, while the rest of values are ignored. In this paper, we propose a new method to estimate the distribution of the complete time series, including both extreme and regular values. This methodology assumes that extreme values time series can be modelled by a normal distribution in a combination of a uniform one. The resulting theoretical distribution is then used to fix the threshold for the POT methodology. The methodology is tested in nine real-world time series collected in the Gulf of Alaska, Puerto Rico and Gibraltar (Spain), which are provided by the National Data Buoy Center (USA) and Puertos del Estado (Spain). By using the Kolmogorov-Smirnov statistical test, the results confirm that the time series can be modelled with this type of mixed distribution. Based on this, the return values and the confidence intervals for wave height in different periods of time are also calculated.

The prediction of convective clouds formation is a very important problem in different areas such as agriculture, natural hazards prevention or transport-related facilities. In this paper, we evaluate the capacity of different types of evolutionary artificial neural networks to predict the formation of convective clouds, tackling the problem as a classification task. We use data from Madrid-Barajas airport, including variables and indices derived from the Madrid-Barajas airport radiosonde station. As objective variable, we use the cloud information contained in the METAR and SPECI meteorological reports from the same airport and we consider a prediction time horizon of 12 h. The performance of different types of evolutionary artificial neural networks has been discussed and analysed, including three types of basis functions (sigmoidal unit, product unit and radial basis function) and two types of models, a mono-objective evolutionary algorithm with two objective functions and a multi-objective evolutionary algorithm optimised by the two objective functions simultaneously. We show that some of the developed neuro-evolutionary models obtain high quality solutions to this problem, due to its high unbalance characteristic.

This work analyses the complementarity and contrast between two metrics commonly used for evaluating the quality of a binary classifier: the correct classification rate or accuracy, C, and the F1 metric, which is very popular when dealing with imbalanced datasets. Based on this analysis, a set of constraints relating C and F1 are defined as a function of the ratio of positive patterns in the dataset. We evaluate the possibility of using a multi-objective evolutionary algorithm guided by this pair of metrics to optimise binary classification models. To check the validity of the constraints, we perform an empirical analysis considering 26 benchmark datasets obtained from the UCI repository and an interesting liver transplant dataset. The results show that the relation is fulfilled and that the use of the algorithm for simultaneously optimising the pair (C,F1) leads to a generally balanced accuracy for both classes. The experiments also reveal that, in some cases, better results are obtained by using the majority class as the positive class instead of using the minority one, which is the most common approach with imbalanced datasets.

The huge amount of data chronologically collected in short periods of time by different devices and technologies is an important challenge in the analysis of times series. This problem has produced the development of new automatic techniques to reduce the number of points in the resulting time series, in order to facilitate their processing and analysis. In this paper, we propose a new modification of a coral reefs optimization algorithm (CRO) to tackle the problem of reducing the size of the time series minimizing the approximation error. The modification includes a memetization procedure (hybridization with a local search procedure) of the standard algorithm to improve its quality when finding a promising solution in a given searching area. The memetization process is applied to the worse individuals of the algorithm at the beginning, and only to the best ones at the end of the algorithm’s convergence, resulting in a dynamical search approach called dynamic memetic CRO (DMCRO). The proposed DMCRO performance is compared in this paper against other state-of-the-art CRO algorithms, such as the standard one, its statistically driven version (SCRO) and two different hybrid versions (HCRO and HSCRO, respectively), and the standard memetic version (MCRO). All the algorithms compared have been tested in 15 time series approximation, collected from different sources, including financial problems, oceanography data, and cardiology signals, among others, showing that the best results are obtained by DMCRO.

Wind power ramp events (WPREs) are strong increases or decreases of wind speed in a short period of time. Predicting WPREs in wind farms is of vital importance given that they can produce damages in the turbines, and, in any case, they suddenly affect the wind farm production. In contrast to previous binary definitions of the prediction problem (ramp vs non-ramp), a three-class prediction model is used in this paper, proposing a novel discretization function, able to detect the nature of WPREs: negative ramp, non-ramp and positive ramp events. Moreover, the natural order of these labels is exploited to obtain better results in the prediction of these events. The independent variables used for prediction include, in this case, past wind speed values and meteorological data obtained from physical models (reanalysis data). Reanalysis will be also used for recovering missing data from the measuring stations in the wind farm. The proposed prediction methodology is based on Reservoir Computing and an over-sampling process for alleviating the high degree of unbalance in the dataset (non-ramp events are much more frequent than ramps). Three elements are combined in the prediction method: a recurrent neural network layer, a nonlinear kernel mapping and an ordinal logistic regression,to exploit the information provided by the order of the classes). Preprocessing is based on a variation of the standard synthetic minority over-sampling technique, which is applied to the reservoir activations (since the direct application over the input variables would damage its temporal structure). The performance of the method is analysed by comparing it against other state-of-the-art classifiers, such as Support Vector Machines, nominal logistic regression, an autoregressive ordinal neural network, or the use of leaky integrator neurons instead of the standard sigmoidal units. From the results obtained, the benefits of the kernel mapping and the ordinal model are clear, and, in general, the performance obtained with the Reservoir Computing approach is shown to be very robust in the detection of ramps.

The discovery of useful patterns embodied in a time series is of fundamental relevance in many real applications. Repetitive structures and common type of segments can also provide very useful information of patterns in financial time series. In this paper, we introduce a time series segmentation and characterization methodology combining a hybrid genetic algorithm and a clustering technique to automatically group common patterns from this kind of financial time series and address the problem of identifying stock market prices trends. This hybrid genetic algorithm includes a local search method aimed to improve the quality of the final solution. The local search algorithm is based on maximizing a likelihood ratio, assuming normality for the series and the subseries in which the original one is segmented. To do so, we select two stock market index time series: IBEX35 Spanish index (closing prices) and a weighted average time series of the IBEX35 (Spanish), BEL20 (Belgian), CAC40 (French) and DAX (German) indexes. These are processed to obtain segments that are mapped into a five dimensional space composed of five statistical measures, with the purpose of grouping them according to their statistical properties. Experimental results show that it is possible to discover homogeneous patterns in both time series.

This paper presents four configurations of a genetic algorithm (GA) combined with a local search (LS) method for time series segmentation with the purpose of correctly recognising extreme values. The LS method is based on likelihood maximisation of a beta distribution. The proposal is tested on three real ocean wave height time series, where extreme values are frequently found. Concretely, the time series are taken from two oceanographic buoys in the Gulf of Alaska, and another one from Puerto Rico. The results show that the different combinations of LS improve the results of the GA. Furthermore, the algorithm provides segmentations where extreme values are grouped in a well-defined cluster, which allows the study of the characteristics of this type of events.

Time series forecasting (TSF) consists on estimating models to predict future values based on previously observed values of time series, and it can be applied to solve many real-world problems. TSF has been traditionally tackled by considering autoregressive neural networks (ARNNs) or recurrent neural networks (RNNs), where hidden nodes are usually configured using additive activation functions, such as sigmoidal functions. ARNNs are based on a short-term memory of the time series in the form of lagged time series values used as inputs, while RNNs include a long-term memory structure. The objective of this paper is twofold. First, it explores the potential of multiplicative nodes for ARNNs, by considering product unit (PU) activation functions, motivated by the fact that PUs are specially useful for modelling highly correlated features, such as the lagged time series values used as inputs for ARNNs. Second, it proposes a new hybrid RNN model based on PUs, by estimating the PU outputs from the combination of a long-term reservoir and the short-term lagged time series values. A complete set of experiments with 29 data sets shows competitive performance for both model proposals, and a set of statistical tests confirms that they achieve the state of the art in TSF, with specially promising results for the proposed hybrid RNN. The experiments in this paper show that the recurrent model is very competitive for relatively large time series, where longer forecast horizons are required, while the autoregressive model is a good selection if the data set is small or if a low computational cost is needed.

Ordinal classification covers those classification tasks where the different labels show an ordering relation, which is related to the nature of the target variable. In addition, if a set of monotonicity constraints between independent and dependent variables has to be satisfied, then the problem is known as monotonic classification. Both issues are of great practical importance in machine learning. Ordinal classification has been widely studied in specialized literature, but monotonic classification has received relatively low attention. In this paper, we define and relate both tasks in a common framework, providing proper descriptions, characteristics, and a categorization of existing approaches in the state-of-the-art. Moreover, research challenges and open issues are discussed, with focus on frequent experimental behaviours and pitfalls, commonly used evaluation measures and the encouragement in devoting substantial research efforts in specific learning paradigms.

16 pages, 5 tables, 3 figures, The analysis of microscopic images of fish gonad cells (oocytes) is a useful tool to estimate parameters of fish reproductive ecology and to analyze fish population dynamics. The study of oocyte dynamics is needed to understand ovary development and reproductive cycle of fish. Oocytes go through different developmental states in a continuum temporal sequence providing an interesting example of ordinal classification, which is not exploited by the current oocyte analysis software. This promising paradigm of machine learning known as ordinal classification or ordinal regression focus on classification problems where there exist a natural order between the classes, thus requiring specific methods and evaluation metrics. In this paper we compare 11 ordinal and 15 nominal state-of-the-art classifiers using oocytes of three fish species (Merluccius merluccius, Trisopterus luscus and Reinhardtius hippoglossoides). The best results are achieved by SVMOD, an ordinal decomposition method of the labelling space based on the Support Vector Machine, varying strongly with the number of states for each specie (about 95 and 80 % of accuracy with three and six states respectively). The classifiers designed specially for ordinal classification are able to capture the underlying nature of the state ordering much better than common nominal classifiers. This is demonstrated by several metrics specially designed to measure misclassification errors associated to states far in the ranking scale., We acknowledge support from the “Junta de Andalucía” under project P11-TIC-7508, from the Spanish Ministry of Science and Innovation (MICINN) under projects TIN2011-22935, TIN2012-32262 and TIN2014-54583-C2-1-R, and from FEDER funds

The mixed use of different shapes of radial basis functions (RBFs) in radial basis functions neural networks (RBFNNs) is investigated in this paper. For this purpose, we propose the use of a generalised version of the standard RBFNN, based on the generalised Gaussian distribution. The generalised radial basis function (GRBF) proposed in this paper is able to reproduce other different radial basis functions (RBFs) by changing a real parameter ?. In the proposed methodology, a hybrid evolutionary algorithm (HEA) is employed to estimate the number of hidden neuron, the centres, type and width of each RBF associated with each radial unit. In order to test the performance of the proposed methodology, an experimental study is presented with 20 datasets from the UCI repository. The GRBF neural network (GRBFNN) was compared to RBFNNs with Gaussian, Cauchy and inverse multiquadratic RBFs in the hidden layer and to other classifiers, including different RBFNN design methods, support vector machines (SVMs), a sparse probabilistic classifier (sparse multinominal logistic regression, SMLR) and other non-sparse (but regularised) probabilistic classifiers (regularised multinominal logistic regression, RMLR). The GRBFNN models were found to be better than the alternative RBFNNs for almost all datasets, producing the highest mean accuracy rank.

Donor---recipient matching constitutes a complex scenario difficult to model. The risk of subjectivity and the likelihood of falling into error must not be underestimated. Computational tools for the decision-making process in liver transplantation can be useful, despite the inherent complexity involved. Therefore, a multi-objective evolutionary algorithm and various techniques to select individuals from the Pareto front are used in this paper to obtain artificial neural network models to aid decision making. Moreover, a combination of two pre-processing methods has been applied to the dataset to offset the existing imbalance. One of them is a resampling method and the other is a outlier deletion method. The best model obtained with these procedures (with AUC = 0.66) give medical experts a probability of graft survival at 3 months after the operation. This probability can help medical experts to achieve the best possible decision without forgetting the principles of fairness, efficiency and equity.

Gene expression detection is a key bioinformatic problem which has been tackled as a classification problem of microarray gene expression, obtained by the light reflection analysis of genomic material. A typical microarray dataset may contain thousands of genes but only a small number of patterns (often less than two hundred). When the dataset presents these kinds of characteristics, state-of-the-art classification models show a high lack of performance. A two-stage algorithm has been proposed to successfully address the problem of microarray classification. In the first stage, two filter algorithms identify salient expression genes from thousands of genes. In the second stage, the proposed methodology is performed using selected gene subsets as new input variables. The methodology proposed is composed of a combination of Logistic Regression (LR) and Evolutionary Generalized Radial Basis Function (EGRBF) neural networks which have shown to be highly accurate in previous research in the modeling of high-dimensional patterns. Finally, the results obtained are contrasted with nonparametric statistical tests and confirm good synergy between EGRBF and LR models.

Combinations of physical and statistical wind speed forecasting models are frequently used in wind speed prediction problems arising in wind farms management. Artificial neural networks can be used in these models as a final step to obtain accurate wind speed predictions. The aim of this work is to determine the potential of evolutionary product unit neural networks (EPUNNs) for improving the accuracy and interpretation of these systems. Traditional neural network and EPUNN approaches have been used to develop different wind speed prediction models. The results obtained using different EPUNN models show that the functional model and the hybrid algorithms proposed provide very accurate prediction compared with standard neural networks used to solve this regression problem. One of the main advantages of the application of these EPUNNs has been the possibility of obtaining some interpretation of the non-linear relation predicted by the model, as will be shown in real data of a wind farm in Spain.

In this paper, a diversity generating mechanism is proposed for an Evolutionary Programming (EP) algorithm that determines the basic structure of Multilayer Perceptron classifiers and simultaneously estimates the coefficients of the models. We apply a modified version of a saw-tooth diversity enhancement mechanism recently presented for Genetic Algorithms, which uses a variable population size and periodic partial reinitializations of the population in the form of a saw-tooth function. Our improvement on this standard scheme consists of guiding saw-tooth reinitializations by considering the variance of the best individuals in the population. The population restarts are performed when the difference of variance between two consecutive generations is lower than a percentage of the previous variance. From the analysis of the results over ten benchmark datasets, it can be concluded that the computational cost of the EP algorithm with a constant population size is reduced by using the original saw-tooth scheme. Moreover, the guided saw-tooth mechanism involves a significantly lower computer time demand than the original scheme. Finally, both saw-tooth schemes do not involve an accuracy decrease and, in general, they obtain a better or similar precision.