Several recent machine learning publications demonstrate the utility of using feature selection algorithms in supervised learning tasks. Among these, scqucnlial feature s1ion algorithms are receiving attention. The most frequently studied variants of these algorithms are forward and backward sequential selection. Many studies on supervised learning with sequential feature selection report applications of these algorithms, but do not consider variants of them that might be more appropriate for some performance tasks. This paper reports positive empirical results on such variants, and argues for their serious consideration in similar learning tasks.

Accurate weather prediction is crucial for many activities, including Naval operations. Researchers within the meteorological division of the Naval Research Laboratory have developed and fielded several expert systems for problems such as fog and turbulence forecasting, and tropical storm movement. They are currently developing an automated system for satellite image interpretation, part of which involves cloud classification. Their cloud classification database contains 204 high-level features, but contains only a few thousand instances. The predictive accuracy of classifiers can be improved on this task by employing a feature selection algorithm. We explain why non-parametric case-based classifiers are excellent choices for use in feature selection algorithms. We then describe a set of such algorithms that use case-based classifiers, empirically compare them, and introduce novel extensions of backward sequential selection that allows it to scale to this task. Several of the approache...

A surprisingly large number of research disciplines have contributed towards the development of knowledge on lazy problem solving, which is characterized by its storage of ground cases and its demand driven response to queries. Case-based reasoning (CBR) is an alternative, increasingly popular approach for designing expert systems that implements this approach. This paper lists pointers to some contributions in some related disciplines that offer insights for CBR research. We then outline a small number of Navy applications based on this approach that demonstrate its breadth of applicability. Finally, we list a few successful and failed attempts to apply CBR, and list some predictions on the future roles of CBR in applications.

The National Severe Storms Laboratory's (NSSL) Mesocyclone Detection Algorithm (MDA) is designed to search for patterns in Doppler velocity radar data which are associated with rotating updrafts in severe thunderstorms. These storm-scale circulations are typically precursors to tornados and severe weather in thunderstorms, yet not all circulations produce such phenomena. A neural network has been designed to diagnose which circulations detected by the NSSL MDA yield tornados. The data used both for the training and the testing of the network is obtained from the NSSL MDA. In particular, 23 variables characterizing the circulations are selected to be used as the input nodes of a feed-forward neural network. The output of the network is chosen to be the existence/nonexistence of tornados, based on ground observations. It is shown that the network outperforms the rule-based algorithm existing in the MDA, as well as statistical techniques such as Discriminant Analysis and Logistic Regressi...

Novel artificial intelligence methods are used to classify 16x16 pixel regions (obtained from Advanced Very High Resolution Radiometer (AVHRR) images) in terms of cloud type (e.g., stratus, cumulus, etc.). We previously reported that intelligent feature selection methods, combined with nearest neighbor classifiers, can dramatically improve classification accuracy on this task. Our subsequent analyses of the confusion matrices revealed that a small number of confusable classes (e.g., cirrus and cirrostratus) dominated the classification errors. We conjectured that, if the class labels in the data were re-represented so that these cloud classes are more easily distinguished, then additional accuracy gains might result. We explored this hypothesis by replacing each class label with a set of error-correcting output codes, a general technique applicable to any classification algorithm for tasks with at least three classes. Our initial results are promising; error correcting codes significa...

Classification problems map an object to a particular class using a set of available features. The problem is how to choose the best subset of features that provide an accurate classification.

A neural network, using input from the Eta Model and upper air soundings, has been developed for the probability of precipitation (PoP) and quantitative precipitation forecast (QPF) for the Dallas–Fort Worth, Texas, area. Forecasts from two years were verified against a network of 36 rain gauges. The resulting forecasts were remarkably sharp, with over 70 % of the PoP forecasts being less than 5 % or greater than 95%. Of the 436 days with forecasts of less than 5 % PoP, no rain occurred on 435 days. On the 111 days with forecasts of greater than 95 % PoP, rain always occurred. The linear correlation between the forecast and observed precipitation amount was 0.95. Equitable threat scores for threshold precipitation amounts from 0.05 in. (�1 mm) to 1 in. (�25 mm) are 0.63 or higher, with maximum values over 0.86. Combining the PoP and QPF products indicates that for very high PoPs, the correlation between the QPF and observations is higher than for lower PoPs. In addition, 61 of the 70 observed rains of at least 0.5 in. (12.7 mm) are associated with PoPs greater than 85%. As a result, the system indicates a potential for more accurate precipitation forecasting.

Ensembles of classifiers have been shown to be very effective for case-based classification tasks. The vast majority of ensemble construction algorithms use the complete set of features available in the problem domain for the ensemble creation. Recent work on randomly selected subspaces for ensemble construction has been shown to improve the accuracy of the ensemble considerably. In this paper we focus our attention on feature selection for ensemble creation using a genetic search approach. We compare boosting and bagging techniques using three approaches for feature selection for ensemble construction.

When dealing with object classification, each object is defined by a set of features (characteristics) that classify the object to a particular class. The problem is how to choose the best subset of characteristics that provide an accurate classification. Previous research has shown that Decision tables are as accurate as C4.5 for classification purposes. Two different genetic search techniques, CHC and CF/RSC, are applied to this problem. Results shows that CF/RSC and Decision tables are a very good combination when dealing with large feature spaces. Results also suggest that CHC is better when used for problems with noise added to the features. 1 Introduction Feature subset selection is defined as a process of selecting a subset of features, d, out of the larger set of D features which maximize the classification performance of a given procedure over all possible subsets [10]. In a typical problem of case-based classification, instances (objects) need to be classified according to ...

this report, we explore the potential of two unsupervised learning programs, AutoClass and K-Means, when applied to a data set that was developed from satellite imagery of cloud regions that were expertly labeled into ten classes. Because cloud types hold meteorological significance, an automated classification from satellite imagery is of obvious use. We compare cloud classes produced by these systems with traditional cloud classes.