Showing papers on "Interpretability published in 2004"

A maximum entropy approach to species distribution modeling

Abstract: We study the problem of modeling species geographic distributions, a critical problem in conservation biology. We propose the use of maximum-entropy techniques for this problem, specifically, sequential-update algorithms that can handle a very large number of features. We describe experiments comparing maxent with a standard distribution-modeling tool, called GARP, on a dataset containing observation data for North American breeding birds. We also study how well maxent performs as a function of the number of training examples and training time, analyze the use of regularization to avoid overfitting when the number of examples is small, and explore the interpretability of models constructed using maxent.

Credit rating analysis with support vector machines and neural networks: a market comparative study

Abstract: Corporate credit rating analysis has attracted lots of research interests in the literature. Recent studies have shown that Artificial Intelligence (AI) methods achieved better performance than traditional statistical methods. This article introduces a relatively new machine learning technique, support vector machines (SVM), to the problem in attempt to provide a model with better explanatory power. We used backpropagation neural network (BNN) as a benchmark and obtained prediction accuracy around 80% for both BNN and SVM methods for the United States and Taiwan markets. However, only slight improvement of SVM was observed. Another direction of the research is to improve the interpretability of the AI-based models. We applied recent research results in neural network model interpretation and obtained relative importance of the input financial variables from the neural network models. Based on these results, we conducted a market comparative analysis on the differences of determining factors in the United States and Taiwan markets.

Classification and Modeling with Linguistic Information Granules: Advanced Approaches to Linguistic Data Mining

Abstract: This book clearly describes how soft computing techniques like fuzzy logic, evolutionary computation and neural networks can be used for extracting interpretable knowledge in the form of linguistic if-then rules from numerical data for classification and modeling. While emphasis is placed on the interpretability of linguistic knowledge, this book covers almost all soft computing techniques for linguistic data mining.

Discovering additive structure in black box functions

Abstract: Many automated learning procedures lack interpretability, operating effectively as a black box: providing a prediction tool but no explanation of the underlying dynamics that drive it. A common approach to interpretation is to plot the dependence of a learned function on one or two predictors. We present a method that seeks not to display the behavior of a function, but to evaluate the importance of non-additive interactions within any set of variables. Should the function be close to a sum of low dimensional components, these components can be viewed and even modeled parametrically. Alternatively, the work here provides an indication of where intrinsically high-dimensional behavior takes place.The calculations used in this paper correspond closely with the functional ANOVA decomposition; a well-developed construction in Statistics. In particular, the proposed score of interaction importance measures the loss associated with the projection of the prediction function onto a space of additive models. The algorithm runs in linear time and we present displays of the output as a graphical model of the function for interpretation purposes.

Design of accurate classifiers with a compact fuzzy-rule base using an evolutionary scatter partition of feature space

Abstract: An evolutionary approach to designing accurate classifiers with a compact fuzzy-rule base using a scatter partition of feature space is proposed, in which all the elements of the fuzzy classifier design problem have been moved in parameters of a complex optimization problem. An intelligent genetic algorithm (IGA) is used to effectively solve the design problem of fuzzy classifiers with many tuning parameters. The merits of the proposed method are threefold: 1) the proposed method has high search ability to efficiently find fuzzy rule-based systems with high fitness values, 2) obtained fuzzy rules have high interpretability, and 3) obtained compact classifiers have high classification accuracy on unseen test patterns. The sensitivity of control parameters of the proposed method is empirically analyzed to show the robustness of the IGA-based method. The performance comparison and statistical analysis of experimental results using ten-fold cross validation show that the IGA-based method without heuristics is efficient in designing accurate and compact fuzzy classifiers using 11 well-known data sets with numerical attribute values.

A Measure of Agreement for Interval or Nominal Multivariate Observations by Different Sets of Judges

Abstract: This article addresses the problem of accounting overall multivariate chance-corrected interobserver agreement when targets have been rated by different sets of judges (not necessarily equal in number). The proposed approach builds on Janson and Olsson’s multivariate generalization of Cohen’s kappa but incorporates weighting for number of judges and applies an expression for expected disagreement suitable for the case with different judges. The authors suggest that the attractiveness of this approach to multivariate agreement measurement lies in the interpretability of the terms of expected and observed disagreement as average distances between observations, and that addressing agreement without regard to the covariance structure among variables has advantages in simplicity and interpretability. Correspondences to earlier approaches are noted, and application of the proposed measure is exemplified using hypothetical data sets.

Simple component analysis

Abstract: Summary. With a large number of variables measuring different aspects of a same theme, we would like to summarize the information in a limited number of components, i.e. linear combinations of the original variables. Among linear dimension reduction techniques, principal component analysis is optimal in at least two ways: principal components extract the maximum of the variability of the original variables, and they are uncorrelated. Unfortunately, they are often difficult to interpret. Moreover, in most applications, only the first principal component is a ‘block component’, the remaining components being ‘difference components’ which are also more difficult to interpret. The goal of simple component analysis is to replace (or to supplement) principal components with suboptimal but better interpretable ‘simple components’. We propose a fast algorithm which seeks the optimal system of components under constraints of simplicity. Thus, in contrast with other techniques like ‘varimax’, this approach always provides a simple solution. The optimal simple system is suboptimal compared with principal components: less variability is extracted and components are correlated. However, if the loss of extracted variability is small, and correlations between components are low, it might be advantageous for practical use. Moreover, our concept of simplicity allows the system to have more than one block component, which also facilitates interpretation. Simplicity is not a guarantee for interpretability. With the help of our algorithm, the user can partly modify an optimal simple system of components to enhance interpretability. In this respect, the ultimate goal of simple component analysis is not to propose a method that leads automatically to a unique solution, but rather to develop tools for assisting the user in his or her choice of an interpretable solution. Finally, we argue that simple components may also make the task of choosing the dimension easier. The methodology is illustrated with a test battery to study the development of neuromotor functions in children and adolescents.

Segmenting Markets by Bagged Clustering

Abstract: We introduce bagged clustering as a new approach in the field of post hoc market segmentation research and illustrate the managerial advantages over both hierarchical and partitioning algorithms, especially with large binary data sets. The most important improvements are enhanced stability and interpretability of segments based on binary data. One of the main goals of the procedure is to complement more traditional techniques as an exploratory segment analysis tool. The merits of the approach are illustrated using a tourism marketing application.

PCA-based feature transformation for classification: issues in medical diagnostics

Abstract: The goal of this paper is to propose, evaluate, and compare several data mining strategies that apply feature transformation for subsequent classification, and to consider their application to medical diagnostics. We (1) briefly consider the necessity of dimensionality reduction and discuss why feature transformation may work better than feature selection for some problems; (2) analyze experimentally whether extraction of new components and replacement of original features by them is better than storing the original features as well; (3) consider how important the use of class information is in the feature extraction process; and (4) discuss some interpretability issues regarding the extracted features.

Heterogeneous fuzzy logic networks: fundamentals and development studies

Abstract: The recent trend in the development of neurofuzzy systems has profoundly emphasized the importance of synergy between the fundamentals of fuzzy sets and neural networks. The resulting frameworks of the neurofuzzy systems took advantage of an array of learning mechanisms primarily originating within the theory of neurocomputing and the use of fuzzy models (predominantly rule-based systems) being well established in the realm of fuzzy sets. Ideally, one can anticipate that neurofuzzy systems should fully exploit the linkages between these two technologies while strongly preserving their evident identities (plasticity or learning abilities to be shared by the transparency and full interpretability of the resulting neurofuzzy constructs). Interestingly, this synergy still becomes a target yet to be satisfied. This study is an attempt to address the fundamental interpretability challenge of neurofuzzy systems. Our underlying conjecture is that the transparency of any neurofuzzy system links directly with the logic fabric of the system so the logic fundamentals of the underlying architecture become of primordial relevance. Having this in mind the development of neurofuzzy models hinges on a collection of logic driven processing units named here fuzzy (logic) neurons. These are conceptually simple logic-oriented elements that come with a well-defined semantics and plasticity. Owing to their diversity, such neurons form essential building blocks of the networks. The study revisits the existing categories of logic neurons, provides with their taxonomy, helps understand their functional features and sheds light on their behavior when being treated as computational components of any neurofuzzy architecture. The two main categories of aggregative and reference neurons are deeply rooted in the fundamental operations encountered in the technology of fuzzy sets (including logic operations, linguistic modifiers, and logic reference operations). The developed heterogeneous networks come with a well-defined semantics and high interpretability (which directly translates into the rule-based representation of the networks). As the network takes advantage of various logic neurons, this imposes an immediate requirement of structural optimization, which in this study is addressed by utilizing various mechanisms of genetic optimization (genetic algorithms). We discuss the development of the networks, elaborate on the interpretation aspects and include a number of illustrative numeric examples.

Model diagnostics for smoothing spline ANOVA models

Abstract: The author proposes some simple diagnostics for assessing the necessity of selected terms in smoothing spline ANOVA models. The elimination of practically insignificant terms generally enhances the interpretability of the estimates and sometimes may also have inferential implications. The diagnostics are derived from Kullback-Leibler geometry and are illustrated in the settings of regression, probability density estimation, and hazard rate estimation.

Making Statistics More Meaningful for Policy Research and Program Evaluation

Abstract: This article focuses on the use of statistics in policy and evaluation research and the need to present statistical information in a form that is meaningful to mixed audiences. Three guidelines for formulating and presenting meaningful statistics are outlined. Understandability ensures that knowledge of statistical methods is not required for comprehending the information presented. Interpretability ensures that statistical information can be explained using familiar, non-abstract units. Comparability ensures that the magnitudes of different estimates can be directly compared within and across studies. Examples for improving popular effect size estimates from linear and non-linear models are included, and a general approach to presenting statistical information meaningfully for consumers of policy and evaluation research is explained.

Synonymous theories in answer set programming and equilibrium logic

Abstract: The study of strong equivalence between logic programs or nonmonotonic theories under answer set semantics, begun in [18], is extended to the case where the programs or theories concerned are formulated in different languages. We suggest that theories in different languages be considered equivalent in the strong sense or synonymous if and only if each is bijectively interpretable (hence translatable) into the other. Since the logic of here-and-there, which provides a suitable foundation for answer set programming, has the Beth property, we can easily give model-theoretic conditions that are equivalent to bijective interpretability. These conditions involve mappings between the models of the two theories that, in particular, preserve the property of being an answer set or equilibrium model.

KBCT: a knowledge extraction and representation tool for fuzzy logic based systems

Abstract: This paper presents a user-friendly portable tool designed and developed in order to make easier knowledge extraction and representation for fuzzy logic based systems. KBCT is an open source software that could be executed under Linux or Windows operating systems. Main goal of KBCT is the generation or refinement of fuzzy knowledge bases with a particular interest of obtaining interpretable partitions and rules. The use of fuzzy logic simplifies the knowledge extraction process and increase interpretability of rules because of the fuzzy rule expression is closed to expert natural language. KBCT lets the user define expert variables and rules, but also provide induction capabilities for partitions and rules. Both types of knowledge, expert and induced, are integrated under the expert control. In addition to this, the user can check consistency and quality of rule base at any moment. A simplify option is implemented in order to allow the user to reduce the size of rule base. The main objective consists of ensuring interpretability, non-redundancy and consistency of the knowledge base along the whole process.

Evolutionary fuzzy modeling human diagnostic decisions.

Abstract: Fuzzy CoCo is a methodology, combining fuzzy logic and evolutionary computation, for constructing systems able to accurately predict the outcome of a human decision-making process, while providing an understandable explanation of the underlying reasoning. Fuzzy logic provides a formal framework for constructing systems exhibiting both good numeric performance (accuracy) and linguistic representation (interpretability). However, fuzzy modeling--meaning the construction of fuzzy systems--is an arduous task, demanding the identification of many parameters. To solve it, we use evolutionary computation techniques (specifically cooperative coevolution), which are widely used to search for adequate solutions in complex spaces. We have successfully applied the algorithm to model the decision processes involved in two breast cancer diagnostic problems, the WBCD problem and the Catalonia mammography interpretation problem, obtaining systems both of high performance and high interpretability. For the Catalonia problem, an evolved system was embedded within a Web-based tool-called COBRA-for aiding radiologists in mammography interpretation.

Cascade Architectures of Fuzzy Neural Networks

Abstract: This study is concerned with cascade architectures of fuzzy neural networks. These networks exhibit three interesting and practically appealing features: (i) come with sound and transparent logic characteristics by being developed with the aid of AND and OR fuzzy neurons and subsequently logic processors (LPs), (ii) possess significant learning abilities and in this way fall in the realm of neuro-fuzzy architectures, and (iii) exhibit an evident hierarchical structure owing to the cascade of the LPs. We discuss main functional properties of the model and relate them to its form of cascade-type of systems formed as a stack of LPs. The construction of the systems of this form calls for some structural optimization that is realized in the realm of genetic optimization. The structure of the network that deals with a selection of a subset of input variables and their distribution across the individual LPs is optimized with the use of genetic algorithms (GAs). The chromosomes encode the order of the variables as well as include the parameters (connections) of the neurons. We discuss various schemes of genetic optimization (both a two-level and single-level GA) and gradient-based learning aimed at further refinement of the connections of the neurons. We elaborate on the interpretation aspects of the network and show how this leads to a Boolean or multivalued logic description of the experimental data. A number of numeric data sets are discussed with respect to the performance of the constructed networks and their interpretability.

FRIwE: fuzzy rule identification with exceptions

Abstract: In this paper, the FRIwE method is proposed to identify fuzzy models from examples. Such a method has been developed trying to achieve a double goal:accuracy and interpretability. In order to do that, maximal structure fuzzy rules are firstly obtained based on a method proposed by Castro et al. In a second stage, the conflicts generated by the maximal rules are solved, thus increasing the model accuracy. The resolution of conflicts are carried out by including exceptions in the rules. This strategy has been identified by psychologists with the learning mechanism employed by the human being, thus improving the model interpretability. Besides, in order to improve the interpretability even more, several methods are presented based on reducing and merging rules and exceptions in the model. The exhaustive use of the training examples gives the method a special suitability for problems with small training sets or high dimensionality. Finally, the method is applied to an example in order to analyze the achievement of the goals.

Recent advances in exploratory data analysis with neuro-fuzzy methods

Abstract: Modern information technology makes it possible today to collect, store, transfer, and combine huge amounts of data at very low costs. Thus more and more companies, and scientific and governmental institutions build up large archives of all kinds of data like numbers, tables, documents, images, sounds, etc. Although users often have a vague understanding of their data and can usually formulate hypotheses and guess dependencies, turning these – often abundantly available – data into useful information turns out to be rather difficult. In response to these challenges a new area of research has emerged, called ‘‘knowledge discovery in databases’’ or ‘‘data mining’’, that tries to provide tools to extract valid, useful, understandable, unknown, and unexpected relationships from large databases [1]. This current form of data analysis is an interdisciplinary field, and employs methods from statistics, soft computing, artificial intelligence and machine learning. The stress lies on the development of techniques that produce human-understandable results, and are suited for large, real world datasets. The data in real world applications has in most cases characteristics that challenge classical analyzing approaches. Besides being heterogeneous – which in its simplest form can mean that we have to deal with numerical and symbolic features – , the data is often of low quality. Algorithms must thus be able to deal with uncertainty and imprecision. The characteristics of the data sources – their quantity, complexity, dimensionality and imperfection – , the essential of extracting understandable patterns from these, and the need to incorporate available background knowledge in that process, makes us assume that (neuro-) fuzzy techniques will play a considerable role in the future of data mining [3]. It is the ability of fuzzy sets to transform between computer representations and (naturally linguistic) human concepts that makes them so valuable to meet the advanced data mining demands, and that Zadeh meant, when he promoted his idea of computing with words [4]. The inherent imprecision of words is not necessarily a weakness, but, on the contrary, can be crucial to model complex systems. From our own experience we observed that many practical applications have this certain robustness where full precision is not necessary. In such cases, exaggerated precision can be a waste of resources, and solutions obtained using fuzzy approaches might be easier to understand and to apply, and gain their strengths by explicitly taking into account vagueness, imprecision or uncertainty. One prominent way to use fuzzy systems in data analysis, is to induce fuzzy if-then rules from data by neurofuzzy techniques. The use of linguistic variables eases the readability and interpretability of the rule base. If we apply such techniques, we must be aware of the trade-off between precision and interpretability. However, the results in data mining are not only judged for their accuracy, but also for their interpretability, as the ultimate goal is to extract human understandable patterns [2]. The papers in this special issue of the Soft Computing Journal deal with quite different aspects of neuro-fuzzy techniques, with a stress on data analysis and rule based systems. Although the intention of this issue cannot be to give a survey of this wide area, it shall give an insight in current trends and research topics in this field. The papers can roughly be divided into three groups:

Improving the interpretability of Takagi-Sugeno fuzzy model by using linguistic modifiers and a multiple objective learning scheme

Abstract: We present a new Tagaki-Sugeno (TS) type model whose membership functions (MFs) are characterized by linguistic modifiers. As a result, during adaptation, the trained local models tend to become the tangents of the global model, leading to good model interpretability. In order to prevent the global approximation ability from being degraded, an index of fuzziness is proposed to evaluate linguistic modification for MFs with adjustable crossover points. A new learning scheme is also developed, which uses the combination of global approximation error and the fuzziness index as its objective function. By minimizing the multiple objective performance measure, a tradeoff between the global approximation and local model interpretation can be achieved. Experimental results show that by the proposed method, good interpretation of local models and transparency of input space partitioning can be obtained for the TS model, while at the same time the global approximation ability is still preserved.

Genetic tuning on fuzzy systems based on the linguistic 2-tuples representation

Abstract: Linguistic fuzzy modeling allows us to deal with the modeling of systems building a linguistic model clearly interpretable by human beings. However, in this kind of modeling the accuracy and the interpretability of the obtained model are contradictory properties directly depending on the learning process and/or the model structure. Thus, the necessity of improving the linguistic model accuracy arises when complex systems are modeled. To solve this problem, one of the research lines of this framework in the last years has leaded up to the objective of giving more accuracy to the linguistic fuzzy modeling, without losing the associated interpretability to a high level. In this work, a new post-processing method of fuzzy rule-based systems is proposed by means of an evolutionary lateral tuning of the linguistic variables, with the main aim of obtaining fuzzy rule-based systems with a better accuracy and maintaining a good interpretability. To do so, this tuning considers a new rule representation scheme by using the linguistic 2-tuples representation model which allows the lateral variation of the involved labels. As an example of application of these kinds of systems, we analyze this approach considering a real-world problem.

Modal matters in interpretability logics

Abstract: In this paper we expose a method for building models for interpretability logics. The method can be compared to the method of taking unions of chains in classical model theory. Many applications of the method share a common part. We isolate this common part in a main lemma. Doing so, many of our results become applications of this main lemma. We also briefly describe how our method can be generalized to modal logics with a different signature. With the general method, we prove completeness for the interpretability logics IL, ILM, ILM0 and ILW*. We also apply our method to obtain a classification of the essential Σ1-sentences of essentially reflexive theories. We briefly comment on such a classification for finitely axiomatizable theories. As a digression we proof some results on self-provers. Towards the end of the paper we concentrate on modal matters con- cerning IL(All), the interpretability logic of all reasonable arithmetical theories. We prove the modal incompleteness of the logic ILW*P0. We put forward a new principle R, and show it to be arithmetically sound in any reasonable arithmetical theory. Finally we make some general remarks on the logics ILRW and IL(All).

Learning maximal structure fuzzy rules with exceptions

Abstract: This paper proposes a method to solve the conflicts that arise in the framework of fuzzy model identification with maximal rules (Fuzzy Sets and Systems 101 (1999) 331) where rules are selected as general as possible. This resolution is expressed by including exceptions in the rules, that way achieving a higher model interpretability with respect to other techniques and a more accurate model. Besides, several methods are presented to improve the interpretability, based on compacting the rules and exceptions of the model. Furthermore, in order to reduce the number of conflicts that arise from the maximal rules, a heuristic strategy is proposed to generate those maximal rules. Finally, the method is applied to an example and the results are compared with other identification methods.

Extracting fuzzy classification rules from partially labeled data

Abstract: The interpretability and flexibility of fuzzy if-then rules make them a popular basis for classifiers. It is common to extract them from a database of examples. However, the data available in many practical applications are often unlabeled, and must be labeled manually by the user or by expensive analyses. The idea of semi-supervised learning is to use as much labeled data as available and try to additionally exploit the information in the unlabeled data. In this paper we describe an approach to learn fuzzy classification rules from partially labeled datasets.

Fuzzy Inference Systems to Model Sensory Evaluation

Abstract: This chapter proposes a fuzzy approach to model the relationship between expert sensory evaluation and consumer preference. An induction method is used to extract qualitative knowledge from the data sample. The induction process is run under interpretability constraints to ensure the fuzzy rules have a meaning for the human expert. To gain interpretability one should tolerate a loss of accuracy. Applying the complete procedure to a food product underlines the importance of data preprocessing and demonstrates that qualitative knowledge can help to relate product attributes to consumer ratings.

Implicit estimation of wiener series

Abstract: The Wiener series is one of the standard methods to systematically characterize the nonlinearity of a system. The classical estimation method of the expansion coefficients via cross-correlation suffers from severe problems that prevents its application to high-dimensional and strongly nonlinear systems. We propose an implicit estimation method based on regression in a reproducing kernel Hubert space that alleviates these problems. Experiments show performance advantages in terms of convergence, interpretability, and system sizes that can be handled

Application of Fuzzy-Integration-Based Multiple-Information Aggregation in Automatic Speech Recognition

Abstract: Many real-world problems can be cast into a multiple-information aggregation framework where preliminary evaluations of separate information sources are combined to produce more accurate and reliable evaluation than would otherwise be the case. In this paper we describe a syllable-proximity evaluation problem in automatic speech recognition that fits well into this aggregation framework. A fuzzy-integration-based approach is adopted as the aggregation operator and a gradient-based algorithm is described for learning parameters automatically from training data. Experiments using spontaneous speech material demonstrate that the fuzzy-integration-based aggregation approach has many advantages over other techniques in terms of both performance and interpretability of the system.