José M. Matías

Explaining and predicting workplace accidents using data-mining techniques

Current research into workplace risk is mainly conducted using conventional descriptive statistics, which, however, fail to properly identify cause-effect relationships and are unable to construct models that could predict accidents. The authors of the present study modelled incidents and accidents in two companies in the mining and construction sectors in order to identify the most important causes of accidents and develop predictive models. Data-mining techniques (decision rules, Bayesian networks, support vector machines and classification trees) were used to model accident and incident data compiled from the mining and construction sectors and obtained in interviews conducted soon after an incident/accident occurred. The results were compared with those for a classical statistical techniques (logistic regression), revealing the superiority of decision rules, classification trees and Bayesian networks in predicting and identifying the factors underlying accidents/incidents.

Reforestation planning using Bayesian networks

The aim of this research was to construct a reforestation model for woodland located in the basin of the river Liebana (NW Spain). This is essentially a pattern recognition problem: the class labels are types of woodland, and the variables for each point are environmental coordinates (referring to altitude, slope, rainfall, lithology, etc.). The model trained using data for existing wooded areas will serve as a guideline for the reforestation of deforested areas. Nonetheless, with a view to tackling reforestation from a more informed perspective, of interest is an interpretable model of relationships existing not just between woodland type and environmental variables but also between and among the environmental variables themselves. For this reason we used Bayesian networks, as a tool that is capable of constructing a causal model of the relationships existing between all the variables represented in the model. The prediction results obtained were compared with those for classical linear techniques, neural networks and support vector machines.

A machine learning methodology for the analysis of workplace accidents

Abstract This article proposes a methodology for the analysis of the causes and types of workplace accidents (in this paper we focus specifically on floor-level falls). The approach is based on machine learning techniques: Bayesian networks trained using different algorithms (with and without a priori information), classification trees, support vector machines and extreme learning machines. The results obtained using the different techniques are compared in terms of explanatory capacity and predictive potential, both factors facilitating the development of risk prevention measures. Bayesian networks are revealed to be the best all-round technique for this type of study, as they combine a powerful interpretative capacity with a predictive capacity that is comparable to that of the best available techniques. Moreover, the Bayesian networks force experts to apply a scientific approach to the construction and progressive enrichment of their models and also enable the basis to be laid for an accident prevention policy that is solidly grounded. Furthermore, the procedure enables better variable definition, better structuring of the data capture, coding, and quality control processes.

Boosting GARCH and neural networks for the prediction of heteroskedastic time series

This work develops and evaluates new algorithms based on GARCH models, neural networks and boosting techniques, designed to model and predict heteroskedastic time series. The main novel elements of these new algorithms are as follows: (a) in regard to neural networks, the simultaneous estimation of the conditional mean and volatility through the maximization of likelihood; (b) in regard to boosting, its simultaneous application to mean and variance components of the likelihood, and the use of likelihood-based models (e.g., GARCH) as the base hypothesis rather than gradient fitting techniques using least squares. The behavior of the proposed algorithms is evaluated over simulated data and over the Standard & Poors 500 Index returns series, resulting in frequent and significant improvements in relation to the ARMA-GARCH models.

Functional classification of ornamental stone using machine learning techniques

Automated classification of granite slabs is a key aspect of the automation of processes in the granite transformation sector. This classification task is currently performed manually on the basis of the subjective opinions of an expert in regard to texture and colour. We describe a classification method based on machine learning techniques fed with spectral information for the rock, supplied in the form of discrete values captured by a suitably parameterized spectrophotometer. The machine learning techniques applied in our research take a functional perspective, with the spectral function smoothed in accordance with the data supplied by the spectrophotometer. On the basis of the results obtained, it can be concluded that the proposed method is suitable for automatically classifying ornamental rock.

Multi-output nonparametric regression

Several non-parametric regression methods with various dependent variables that are possibly related are explored. The techniques which produce the best results in the simulations are those which incorporate the observations of the other response variables in the estimator. Compared to analogous single-response techniques, this approach results in a significant reduction in the quadratic error in the response.

Machine learning techniques applied to the determination of osteoporosis incidence in post-menopausal women

Osteoporosis is a disease that mostly affects women in developed countries. It is characterised by reduced bone mineral density (BMD) and results in a higher incidence of fractured or broken bones. In this research we studied the relationship between BMD and diet and lifestyle habits for a sample of 305 post-menopausal women by constructing a non-linear model using the regression support vector machines technique. One aim of this model was to make an initial preliminary estimate of BMD in the studied women (on the basis of a questionnaire with questions mostly on dietary habits) so as to determine whether they needed densitometry testing. A second aim was to determine the factors with the greatest bearing on BMD with a view to proposing dietary and lifestyle improvements. These factors were determined using regression trees applied to the support vector machines predictions.

Functional support vector machines and generalized linear models for glacier geomorphology analysis

We propose a functional pattern recognition approach to the problem of identifying the topographic profiles of glacial and fluvial valleys, using a functional version of support vector machines (SVMs) for classification. We compare a proposed functional version of SVMs with functional generalized linear models and their vectorial versions: generalized linear models and SVMs that use the original observations as input. The results indicate the benefit of our proposed functional SVMs and, in more general terms, the advantages of using a functional rather than a vectorial approach.

Functional Pattern Recognition of 3D Laser Scanned Images of Wood-Pulp Chips

We evaluate the appropriateness of applying a functional rather than the typical vectorial approach to a pattern recognition problem. The problem to be resolved was to construct an online system for controlling wood-pulp chip granulometry quality for implementation in a wood-pulp factory. A functional linear model and a functional logistic model were used to classify the hourly empirical distributions of wood-chip thicknesses estimated on the basis of images produced by a 3D laser scanner. The results obtained using these functional techniques were compared to the results of their vectorial counterparts and support vector machines, whose input consisted of several statistics of the hourly empirical distribution. We conclude that the empirical distributions have sufficiently rich functional traits so as to permit the pattern recognition process to benefit from the functional representation.

Functional experiment design for the analysis of colour changes in granite using new L * a * b * functional colour coordinates

We propose a functional data approach to evaluating colour changes in stone that is based on applying a functional experiment design to the tristimulus curves resulting from the product of the power spectral distribution of the source, the stone reflectance curve and the matching colour functions of the standard observer. The proposed method was applied to an analysis of colour changes in granite after the application of different desalination treatments. The results were compared with those obtained by the classical analysis of variance applied to the colorimetric coordinates L^*a^*b^*. The granite RGB and XYZ colour coordinate systems were obtained by integrating the respective tristimulus curves. The L^*a^*b^* coordinates, however, were obtained directly by transforming the XYZ coordinates, as no corresponding tristimulus functions have been proposed to date. With a view to comparing the results for these functional and scalar methods for a uniform colour measurement system, these functions, whose integral coincides with the L^*a^*b^* values, have been deduced and proposed for the first time. The results obtained demonstrate the usefulness of the additional information supplied by the functional approach. However, this information does not replace that produced by the scalar approach for the scalar coordinates, and so it is recommended to use both approaches. The new tristimulus functions associated with the L^*a^*b^* coordinates are perfectly interpretable in a way analogous to the coordinates themselves, i.e., as the degree of luminosity (L^*), the green-red relative position (a^*) and the blue-yellow relative position (b^*), except that they are interpreted for each infinitesimal wavelength interval. A brief introduction to the colour measurement problem and to functional statistical techniques is provided for readers coming from different disciplines.