The difficulty of understanding a financial institution's risk of default has been highlighted by multiple recent episodes in both the U.S. and in Europe. This paper describes a study on the empirical comparison of classification techniques for predictive ranking of the 12 month risk of default in banks. This work compares the scoring capabilities of different predictive models. The models compared were induced from past levels of risk exposure observed in historic data. The ranking performance of the models is compared by assessing the highest risk cases, using the left-hand side of the model's ROC curves (i.e., curves representing true positive to false positive rates). Empirical comparisons were performed using FDIC call report data and a one-year-ahead ranking prediction schema. This comparison demonstrates that inductive machine learning techniques can be successfully applied for predictive ranking of default risk. Observed results indicate better performance by symbolic rule or decision tree based models than by traditional modeling techniques based on statistical algorithms.

The Marine Corps' Project Albert seeks to model complex phenomenon by observing the behavior of relatively simple simulations over thousands of runs. These simulations are based upon lightweight agents, whose essential behavior has been distilled down to a small number of rules. By varying the parameters of these rules, Project Albert simulations can explore emergent complex nonlinear behaviors with the aim of developing insight not readily provided by first principle mathematical models. Thousands of runs of Albert simulation models create large amount of data that describe the association/correlation between the simulation input and output parameters. Understanding the associations between the simulation input and output parameters is critical to understanding the simulated complex phenomenon. This paper presents a data mining approach to analyzing the large scale and highly uncertain Albert simulation data. Specifically, a characteristic rule discovery algorithm is described in the paper together with its application to the Albert simulation runtime data.

We address the problem of crafting visual routines for detection tasks. Emphasis is placed on both competition and learning to help with specific visual tasks involved in localization and identification. Crafting of visual routines presents difficult optimization problems and leads to evolutionary computation using a hybrid genetic architecture consisting of natural selection, learning, and their beneficial interactions. Base features representations and visual routines for detection represented as decision trees are evolved. The visual routine considered is that of eye detection. The experimental results reported herein prove the feasibility of our approach in terms of feature selection (data compression) and the corresponding eye detection (pattern recognition).

This paper describes a hybrid methodology that integrates genetic algorithms (GAs) and decision tree learning in order to evolve useful subsets of discriminatory features for recognizing complex visual concepts. A GA is used to search the space of all possible subsets of a large set of candidate discrimination features. Candidate feature subsets are evaluated by using C4.5, a decision tree learning algorithm, to produce a decision tree based on the given features using a limited amount of training data. The classification performance of the resulting decision tree on unseen testing data is used as the fitness of the underlying feature subset. Experimental results are presented to show how increasing the amount of learning significantly improves feature set evolution for difficult visual recognition problems involving satellite and facial image data. In addition, we also report on the extent to which other more subtle aspects of the Baldwin effect are exhibited by the system.

This paper describes how to integrate subsymbolic and symbolic processes in order to create high-performance shape analysis systems. The specific methodology introduced integrates morphological processing and machine learning techniques such as genetic algorithms (GAs) and empirical inductive generalization. The optimal operators (defined as variable morphological structuring elements) evolved by GAs are used to derive discriminant feature vectors, which are then used by empirical inductive learning to generate rule-based class description in disjunctive normal form. The rule-based descriptions are finally optimized by removing small disjuncts in order to enhance the robustness of the shape analysis system. Experimental results are presented to illustrate the feasibility of the methodology for discriminating among classes of arbitrarily shaped objects, for learning the concepts of convexity and concavity, and for building robust recognition methods.>

An iterative noise reduction learning algorithm is presented in which rules are learned in two phases. The first phase improves the quality of training data through a concept-driven closed-loop filtration process. In the second phase, classification rules are relearned from the filtered training data set.>

This paper presents an approach to the recognition of noisy patterns. We assume that learned concept descriptions contain noisy components. These descriptions can be optimized to gain better performance through the elimination of some less significant noisy components. Parametric control of such an optimization, however, is very difficult and a parameter value has to be found through the training process. This value can change over concept variations and time, and in consequence, it can decrease the system recognition effectiveness. We found that the classification decision can be based on the dynamic characteristics of the recognition curve acquired over increasing degree of optimization. In this approach, the optimization and matching steps are repeated iteratively. The dynamic characteristics of recognition curves are completed and a pattern of each recognition curve is labeled. For a given set of test data, the system selects recognition curves of uptrend patterns (when the recognition rate increases with the increase of the optimization degree), and the final classification is made for the class which has the highest recognition rates among selected uptrend recognition curves. Thus, the method classifies data based on a series of matches that determines the behavior of the recognition curve instead of the traditional single match. The effectiveness of the proposed methodology is illustrated on the example of specially prepared very noisy texture attributes.

The TEXTRAL system, used for determining structural visual properties of textures through symbolic transformations, is presented. The method consists of two phases: one that extracts information from raw textural images by applying convolutional operators and learns an initial set of rules; and a second that iteratively extracts symbolic information from the transformed representation of initial image and learns another set of rules. The transformed symbolic representation is obtained by applying previously learned rules to a new image location and generating symbolic images based on rule assertions.>

A novel way of combining morphological processing and genetic algorithms (GAs) to generate high-performance shape discrimination operators is presented. GAs can evolve operators that discriminate among classes comprising different shapes. The operators are defined as variable structuring elements and can be sequenced as program forms. The population of such operators, evaluated according to an index of performance corresponding to shape discrimination ability, evolves into an optimal set of operators using genetic search. Experimental results are presented to illustrate the feasibility of the approach for shape discrimination.>