Elena Montañés

Introducing a family of linear measures for feature selection in text categorization

Text categorization, which consists of automatically assigning documents to a set of categories, usually involves the management of a huge number of features. Most of them are irrelevant and others introduce noise which could mislead the classifiers. Thus, feature reduction is often performed in order to increase the efficiency and effectiveness of the classification. In this paper, we propose to select relevant features by means of a family of linear filtering measures which are simpler than the usual measures applied for this purpose. We carry out experiments over two different corpora and find that the proposed measures perform better than the existing ones.

Dependent binary relevance models for multi-label classification

Several meta-learning techniques for multi-label classification (MLC), such as chaining and stacking, have already been proposed in the literature, mostly aimed at improving predictive accuracy through the exploitation of label dependencies. In this paper, we propose another technique of that kind, called dependent binary relevance (DBR) learning. DBR combines properties of both, chaining and stacking. We provide a careful analysis of the relationship between these and other techniques, specifically focusing on the underlying dependency structure and the type of training data used for model construction. Moreover, we offer an extensive empirical evaluation, in which we compare different techniques on MLC benchmark data. Our experiments provide evidence for the good performance of DBR in terms of several evaluation measures that are commonly used in MLC. HighlightsWe propose DBR as a multi-label classifier that exploits conditional label dependence.DBR combines properties of both, chaining and stacking learning strategies.We provide a careful analysis of the relationship between these techniques.We study the underlying dependency structure and the type of training data used.Our experiments show the good performance of DBR in terms of several measures.

Improving performance of text categorization by combining filtering and support vector machines

Text Categorization is the process of assigning documents to a set of previously fixed categories. A lot of research is going on with the goal of automating this time-consuming task. Several different algorithms have been applied, and Support Vector Machines (SVM) have shown very good results. In this report, we try to prove that a previous filtering of the words used by SVM in the classification can improve the overall performance. This hypothesis is systematically tested with three different measures of word relevance, on two different corpus (one of them considered in three different splits), and with both local and global vocabularies. The results show that filtering significantly improves the recall of the method, and that also has the effect of significantly improving the overall performance.

Analysis of the thermal performance of a church window steam condenser for different operational conditions using three models

Abstract The results of the usual design model “Standards for Steam Surface Condensers (SSSC)” and those of a two-dimensional model (2D) and a three-dimensional model (3D) are presented. Both the 2D model and the 3D model allow the presence of non-condensable gases, the pressure drop across the tube bundle and the tube arrangement to be taken into account, while the SSSC model does not. In order to study the thermal performance of the models, the heat transfer rate and the condensate mass flow rate were analysed varying the circulating water inlet temperature and velocity, as well as the fouling inside the tubes. The 2D and 3D models behave similarly to the SSSC model, but present less dependence on temperature and velocity variations. Because of the low running time, both the SSSC and the 2D models are adequate for on-line fouling evaluation of the condenser, the latter affording a more detailed description of the condenser. The SSSC model and the 3D model can complement one another in the design stage, using the SSSC for the area estimation and improving the inner condenser geometry by means of the 3D model.

Enhancing directed binary trees for multi-class classification

One approach to multi-class classification consists in decomposing the original problem into a collection of binary classification tasks. The outputs of these binary classifiers are combined to produce a single prediction. Winner-takes-all, max-wins and tree voting schemes are the most popular methods for this purpose. However, tree schemes can deliver faster predictions because they need to evaluate less binary models. Despite previous conclusions reported in the literature, this paper shows that their performance depends on the organization of the tree scheme, i.e. the positions where each pairwise classifier is placed on the graph. Different metrics are studied for this purpose, proposing a new one that considers the precision and the complexity of each pairwise model, what makes the method to be classifier-dependent. The study is performed using Support Vector Machines (SVMs) as base classifiers, but it could be extended to other kind of binary classifiers. The proposed method, tested on benchmark data sets and on one real-world application, is able to improve the accuracy of other decomposition multi-class classifiers, producing even faster predictions.

Aggregating independent and dependent models to learn multi-label classifiers

The aim of multi-label classification is to automatically obtain models able to tag objects with the labels that better describe them. Despite it could seem like any other classification task, it is widely known that exploiting the presence of certain correlations between labels helps to improve the classification performance. In other words, object descriptions are usually not enough to induce good models, also label information must be taken into account. This paper presents an aggregated approach that combines two groups of classifiers, one assuming independence between labels, and the other considering fully conditional dependence among them. The framework proposed here can be applied not only for multi-label classification, but also in multi-label ranking tasks. Experiments carried out over several datasets endorse the superiority of our approach with regard to other methods in terms of some evaluation measures, keeping competitiveness in terms of others.

A wrapper approach with support vector machines for text categorization

The ScanningN-T uple classifier (SNT) was introduced by Lucas and Amiri [1, 2] as an efficient and accurate classifier for chain-coded hand-written digits. The SNT operates as speeds of tens of thousands of sequences per second, during both the trainingand the recognition phases. The main contribution of this paper is to present a new discriminative trainingrule for the SNT. Two versions of the rule are provided, based on minimizingthe mean-squared error and the cross-entropy, respectively. The discriminative trainingrule offers improved accuracy at the cost of slower trainingtime, since the trainingis now iterative instead of single pass. The cross-entropy trained SNT offers the best results, with an error rate of 2.5% on sequences derived from the MNIST test set.

Measures of Rule Quality for Feature Selection in Text Categorization

Text Categorization is the process of assigning documents to a set of previously fixed categories. A lot of research is going on with the goal of automating this time-consuming task. Several different algorithms have been applied, and Support Vector Machines have shown very good results. In this paper we propose a new family of measures taken from the Machine Learning environment to apply them to feature reduction task. The experiments are performed on two different corpus (Reuters and Ohsumed). The results show that the new family of measures performs better than the traditional Information Theory measures.

Forecasting Time Series Combining Machine Learning and Box-Jenkins Time Series

In statistics, Box-Jenkins Time Series is a linear method widely used to forecasting. The linearity makes the method inadequate to forecast real time series, which could present irregular behavior. On the other hand, in artificial intelligence FeedForward Artificial Neural Networks and Continuous Machine Learning Systems are robust handlers of data in the sense that they are able to reproduce nonlinear relationships. Their main disadvantage is the selection of adequate inputs or attributes better related with the output or category. In this paper, we present a methodology that employs Box-Jenkins Time Series as feature selector to Feedforward Artificial Neural Networks inputs and Continuous Machine Learning Systems attributes. We also apply this methodology to forecast some real time series collected in a power plant. It is shown that Feedforward Artificial Neural Networks performs better than Continuous Machine Learning Systems, which in turn performs better than Box-Jenkins Time Series.

Improving the Discriminatory Power of a Near-Infrared Microscopy Spectral Library with a Support Vector Machine Classifier

A multi-group classifier based on the support vector machine (SVM) has been developed for use with a library of 48 456 spectra measured by near-infrared reflection microscopy (NIRM) on 227 samples representing 26 animal feed ingredients and 4 possible contaminants of animal origin. The performance of the classifier was assessed by a five-fold cross-validation, dividing at the sample level. Although the overall proportion of misclassifications was 27%, almost all of these involved the confusion of pairs of similar ingredients of vegetable origin. Such confusions are unimportant in the context of the intended use of the library, which is the detection of banned ingredients in animal feed. The error rate in discrimination between permitted and banned ingredients was just 0.17%. The performance of the SVM classifier was substantially better than that of the K-nearest-neighbors method employed in previous work with the same library, for which the comparable error rates are 36% overall and 0.39% for permitted versus banned ingredients.