This paper explores approaches to sentiment classification of U.S. Congressional floordebate transcripts. Collective classification techniques are used to take advantage of the informal citation structure present in the debates. We use a range of methods based on local and global formulations and introduce novel approaches for incorporating the outputs of machine learners into collective classification algorithms. Our experimental evaluation shows that the mean-field algorithm obtains the best results for the task, significantly outperforming the benchmark technique. 1

Abstract. This paper investigates the use of machine learning classification techniques applied to the task of recognising the make and model of vehicles. Although a number of vehicle classification systems already exist, most of them seek only to distinguish between vehicle categories, e.g. identifying whether a vehicle is a bus, truck or car. The system presented here demonstrates that a set of features extracted from the frontal view of a vehicle may be used to determine the vehicle type (make and model) with high accuracy. The performance of some standard multi-class classification algorithms is compared for this problem. A one-class k-Nearest Neighbour classification algorithm is also implemented and tested. 1

Abstract—In this paper, we face the problem of designing accurate decision-making modules in measurement systems that need to be implemented on resource-constrained platforms. We propose a methodology based on multiobjective optimization and genetic algorithms (GAs) for the analysis of support vector machine (SVM) solutions in the classification error-complexity space. Specific criteria for the choice of optimal SVM classifiers and experimental results on both real and synthetic data will also be discussed. Index Terms—Classification accuracy, genetic algorithms (GAs), learning-from-examples classifiers, multiobjective optimization (MOO). I.

Abstract—The design of low-cost distributed real-time classifiers whose inputs are the physical data from the environment is an issue of major interest in the emerging technology of socalled smart sensors. When a classifier has to be implemented on low-power and low-cost platforms, a tradeoff between classification accuracy and implementation complexity must be pursued. Here, a multiobjective optimization approach will be introduced to jointly minimize both the classification error rate and the platform resource usage. Objective evaluation is then a critical issue because design decision making is based on that. In practice, objective estimation is usually affected by uncertainty, which has to be taken into account in the design process. Here, we will consider the uncertainty that originates from the reduced size of the manually classified (labeled) data sets, which form the sole source of information used to build a learning-from-examples classifier. Then, the design criteria that make direct use and even try to take advantage of such uncertainty will be proposed. The proposed approach is validated using both synthetic and real-world data sets. Index Terms—Estimation uncertainty, learning-from-examples classifiers, multiobjective optimization (MOO), resource constrained

Often, the explanatory power of a learned model must be traded off against model performance. In the case of predicting runaway software projects, we show that the twin goals of high performance and good explanatory power are achievable after applying a variety of data mining techniques (discrimination, feature subset selection, rule covering algorithms). This result is a new high water mark in predicting runaway projects. Measured in terms of precision, this new model is as good as can be expected for our data. Other methods might out-perform our result (e.g. by generating a smaller, more explainable model) but no other method could out-perform the precision of our learned model.

learning algorithms using sign tests

The classification of new cases using a predictive model incurs two types of costs—testing costs and misclassification costs. Recent research efforts have resulted in several novel algorithms that attempt to produce learners that simultaneously minimize both types. In many real life scenarios, however, we cannot afford to conduct all the tests required by the predictive model. For example, a medical center might have a fixed predetermined budget for diagnosing each patient. For cost bounded classification, decision trees are considered attractive as they measure only the tests along a single path. In this work we present an anytime framework for producing decision-tree based classifiers that can make accurate decisions within a strict bound on testing costs. These bounds can be known to the learner, known to the classifier but not to the learner, or not predetermined. Extensive experiments with a variety of datasets show that our proposed framework produces trees with lower misclassification costs along a wide range of testing cost bounds.