This study compares eight international failure prediction models on one data set of Belgian company accounts, using performance indicators based on the inequality principle and performance measures based on a classification rule. After a brief theoretical review of the two basic modelling techniques in failure prediction research and the performance measures used to evaluate them, we report type I and type II error rates corresponding with the original cut-off point and calculate new optimal cut-off points, as well as Gini-coefficients. A wide range of performances was observed for the different models. However the models estimated on a sample of Continental European companies are found to be better performing when validated on a sample of Continental European, i.e. Belgian companies, than the Anglo-Saxon models. A remarkable finding is also that the Greek Gloubos-Grammaticos models show better predictive ability when validated on samples of Belgian failing and non-failing companies than on their own (Greek) validation samples. Another important finding is the robustness of the older discriminant models and the models that were estimated on bigger companies. The validation shows that very simple models can have great predictive ability. 2

Faced with the question whether the Belgian failure prediction models by Ooghe, Joos and De Vos (1991) can be easily applied in all industries and for all sizeclasses, this study compares the performance of the OJD models across 18 different industries and different sizeclasses. After a brief theoretical review of the logistic regression modelling technique, which was used to design the OJD 1991 models, and the performance measures that are used to evaluate these models, we report type I and type II error rates corresponding with the original cut-off points of the models. Furthermore, we calculate new optimal cut-off points, as well as Ginicoefficients. Finally we report the reductions in unweighted error rates when using the new cut-off points instead of the original ones, and the graphs of the trade-off functions. As can be concluded from the performance results and the trade-off functions, there’s a wide range of performances for the different industries. However, we notice that the OJD 1991 models perform best for classical manufacturing industries- such as chemicals, paper and printing, textiles and apparel, paper and printing and metal- and financial services., while the models show the worst performance for service industries- such as real estate, hotel,

The traditional rules of voice-leading in Western music are explicated using experimentally established perceptual principles. Six core principles are shown to account for the majority of voice-leading rules given in historical and contemporary music theory tracts. These principles are treated in a manner akin to axioms in a formal system from which the traditional rules of voice-leading are derived. Nontraditional rules arising from the derivation are shown to predict formerly unnoticed aspects of voice-leading practice. In addition to the core perceptual principles, several auxiliary principles are described. These auxiliary principles are occasionally linked to voice-leading practice and may be regarded as compositional “options ” that shape the music-making in perceptually unique ways. It is suggested that these auxiliary principles distinguish different types of part writing, such as polyphony, homophony, and close harmony. A theory is proposed to account for the aesthetic origin of voiceleading practices.

: We develop a proximity function between an individual and a population from a distance between multivariate observations. We study some properties of this construction and apply it to a distance--based discrimination rule, which contains the classic linear discriminant function as a particular case. Additionally, this rule can be used advantageously for categorical or mixed variables, or in problems where a probabilistic model is not well determined. This approach is illustrated and compared with other classic procedures using four real data sets. Keywords: Categorical and mixed data; Distances between observations; Multidimensional scaling; Discrimination; Classification rules. AMS Subject Classification: 62H30 The authors thank M.Abrahamowicz, J. C. Gower and M. Greenacre for their helpful comments, and W. J. Krzanowski for providing us with a data set and his quadratic location model program. Work supported in part by CGYCIT grant PB93--0784. Authors' address: Departam...

We propose a framework and a complete implementation of a translation and scale invariant image recognition system for natural indoor scenes. The system employs higher order autocorrelation features of scale space data which permit linear classification. An optimal linear classification method is presented, which is able to cope with a large number of classes represented by many, as well as very few samples. In the course of the analysis of our system, we examine which numerical methods for feature transformation and classification show sufficient stability to fulfill these demands. The implementation has been extensively tested. We present the results of our own application and several classification benchmarks. Image recognition Face recognition Scale invariancy Scale space Higher order autocorrelation Optimal linear classification 1. INTRODUCTION The task of visual recognition which was defined by Marr (1) with the question: "What objects are where in the environment?" is still ...

Twelve male listeners categorized 54 synthetic vowel stimuli that varied orthogonally in F2 and F3 on a BARK scale into the American English vowel categories /I/, /U/, and / ˛ /. A neuropsychological model of visual categorization, called the Striatal Pattern Classifier (SPC; [1]) is generalized to the auditory domain, and applied separately to the data from each observer. Performance of the SPC is compared with the successful Normal A Posteriori Probability model (NAPP; [2], [3]) of auditory categorization. Versions of the SPC and NAPP that assume linear response region partitions provided similar accounts of the data. Nonlinear versions of both models provided only small improvements in fit. 1.

This paper examines the efficiency of a generalized adaptive neural network algorithm (GANNA) processor in comparison to earlier model based methods, a back propagation artificial neural network, and logistic regression approaches to data classification. The research uses the binary classification problem of discriminating between failing and non-failing firms to compare the methods. The results indicate the potential in time savings and the successful classification results available from a GANNA processor.

Abstract. The recognition of affective human communication may be used to provide developers with a rich source of information for creating systems that are capable of interacting well with humans. Posture has been acknowledged as an important modality of affective communication in many fields. Behavioral studies have shown that posture can communicate discrete emotion categories as well as affective dimensions. In the affective computing field, while models for the automatic recognition of discrete emotion categories from posture have been proposed, to our knowledge, there are no models for the automatic recognition of affective dimensions from static posture. As a continuation of our previous study, the two main goals of this study are: i) to build automatic recognition models to discriminate between levels of affective dimensions based on low-level postural features; and ii) to investigate both the discriminative power and the limitations of the postural features proposed. The models were built on the basis of human observers ’ ratings of posture according to affective dimensions directly (instead of emotion category) in conjunction with our posture features. 1

. In this paper we use canonical discriminant analysis and principal component analysis to analyze literary styles and distinguish authorship. The use of these techniques in conjunction with statistical graphics reduces the need for imposing unrealistic distributional assumptions on the data while still retaining the ability to separate works into authorship groups. We use counts of function words as our units of analysis and find this approach to be e#ective. 1. Introduction The analysis of literary style is a field with many opportunities for statisticians. The increasing amount of data available allows for the testing and application of various statistical techniques. Since the problems involved in analyzing literary style are inherently multivariate, classical techniques such as canonical discriminant analysis and principal component analysis can be used. Although it is impossible to identify all of the di#erent factors that go into the writing of a piece of literature, one goal i...

This paper describes an application of an inductive learning techniques to case-based reasoning. We introduce two main forms of induction, define case-based reasoning and present a combination of both. The evaluation of the proposed system, called TA3, is carried out on a classification task, namely character recognition. We show how inductive knowledge improves knowledge representation and in turn flexibility of the system, its performance (in terms of classification accuracy) and its scalability. 1. Introduction Inductive learning is a process of generalizing specific facts or observations [MCM86]. It is a basic strategy by which one can acquire knowledge. There are two main forms associated with inductive learning: 1. Instance-to-class induction, where the learning system is presented with independent instances, representing class and the task is to induce a general description of the class. 2. Clustering problem arises when several objects or situations are presented to a learner...