L. Breiman, J. H. Friedman, R. A. Ohlsen, C. J. Stone: Classification And Regression Trees. Wadsworth, Belmont, CA, 1984. 643, 644  Home/Search   Document Not in Database   Summary   Related Articles   Check

....context, the prior probability of a risk class Low High C , is given by the prevalence h p . For every observation i in h D the posterior probability of being in a High risk class is denoted by i p and of being in a Low risk class by i p 1 . Some classifiers, such as decision trees [12][48] return a subset at the leaves and take the most frequent class as the predicted value. However, each leaf has a distribution and therefore i p can take on any value between 0 and 1. Other classifiers, such as k nearest neighbor with k=1, return a single similar observation from the case ....

L. Breiman, J. Friedman, R. Olshen, and C. Stone: Classification and Regression Trees. Wadsworth, 1984.

....of interdependencies are to be considered. Because it is a difficult task to try all combinations of ranges of (input) metrics and determine their individual influence on the classification algorithm to predict quality factors, such automated techniques have been developed that solve this problem [9,19,20]. Pareto Classification Pareto analysis is included as a classification technique that is common for quick quality analyses. The goal of a Pareto analysis is to identify those 20 of 10 all components that contribute heavily to all troubles. The principle is nicknamed 80:20 rule because it ....

.... effects of uncorrelated metrics in input training data (0, 0 robustness to outlying data sets during training (0, 0 portability to data sets from other projects with same design methodology (0, 0 bibliography for applications and theory [2,9,19,23] [4,21] 17] 12,15, 18,22] dependent on learning approach or classification algorithm 0 bad results medium results good results 24 VIII. Summary and Further Research We have evaluated several classification techniques as an approach for predicting faults based on code complexity ....

Breiman, L., J.H.Friedman, R.A.Olshen, and C.J.Stone: Classification and Regression Trees. Wadsworth, Belmont, CA, 1984.

....of numerical attributes and mappings of such attributes is a well known task in statistics. Such a mapping is usually approximated by a regression function of an appropriate type. Traditional statistical regression techniques have been generalized to so called regression trees; the algorithm CART [3] is one of such pioneers in this field. Most extensions and modifications of ML algorithms that process continuous classes have been so far done for TDIDT algorithms that induce decision trees; namely RETIS [6] and M5 [8] FORS [7] is one of the first attempt to incorporate continuous classes to ....

L. Breiman, J.H. Friedman, R.A. Olshen, and C.J. Stone: Classification and Regression Trees. Wadsworth Int. Group, Belmont, California 1984.

....way for performing this reduction. Note that we are following an exploratory strategy here, as opposed to a confirmatory one since the association structure is not stipulated beforehand. The algorithm that we used to construct classification trees was CART (Classification and Regression Trees) [1]. Constructing classification trees with CART requires that the dependent variable be dichotomized. We did so around the median value, differentiating between low success organizations and high success organizations. Classification tree algorithms have a number of analytical advantages. ....

....possible positions to make a split. There are a number of different ways in which the goodness of a split can be judged. In practice, the literature suggests that not much difference exists between the commonly used splitting criteria in terms of the accuracy of the tree that is constructed [17][1]. The splitting criterion we use is the Gini measure of node impurity [1] The Gini measure reaches a value of zero when only one dependent variable class is present at a node. This measure is computed as the sum of products of all pairs of class proportions for classes present at the node. It ....

[Article contains additional citation context not shown here]

L. Breiman, J. Friedman, R. Olshen, and C. Stone: Classification and Regression Trees. Wadsworth, 1984.

....any other function estimating the goodness of split of decision trees could be used as evaluation function. It is crucial to consider that the operator performs hill climbing and is therefore generally not able to find globally optimal solutions. In contrast to o , an operator introduced in [Breiman et al. 1984] is capable of finding optimal solutions. There are two major differences to be noted here: Although Breiman s operator could easily be generalized to make it applicable to any number of operands (like o ) it was originally applied to a single operand. Secondly, Breiman s operator only ....

L. Breiman, J. Friedman, R. Olshen, and C. Stone: Classification and Regression Trees. Wadsworth & Brooks, Pacific Grove, CA, 1984.

....An extension of this approach that is more applicable to small samples is to use N fold 7 As a discipline, it is important for software engineering to develop general measures. However, the extent to which this can be achieved in practice remains an empirical question. cross validation [B84]. If the sample is of size N, then this approach stipulates that the analysis be performed repeatedly for each n, n=1, N, with n removed from the sample and using the remaining N 1 observations. When ordinary least squares regression is utlized for validation, for example, the PRESS ....

L. Breiman, J. Friedman, R. Olshen, and C. Stone: Classification and Regression Trees, Wadsworth, 1984.

No context found.

L. Breiman, J. H. Friedman, R. A. Ohlsen, C. J. Stone: Classification And Regression Trees. Wadsworth, Belmont, CA, 1984. 643, 644

No context found.

Leo Breiman, J.H. Friedman, R.A.Olshen, C.J. Stone: Classification and regression trees, Wadsworth 1984

No context found.

Leo Breiman, J.H. Friedman, R.A.Olshen, C.J. Stone: Classification and regression trees, Wadsworth 1984

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC