BACKGROUND OF THE INVENTION

[0001]
1. Field of the Invention

[0002]
The present invention generally relates to the field of costsensitive learning in the areas of machine learning and data mining and, more particularly, to methods for solving multiclass costsensitive learning problems using a binary classification algorithm. This algorithm is based on techniques of data space expansion and gradient boosting with stochastic ensembles.

[0003]
2. Background Description

[0004]
Classification in the presence of varying costs associated with different types of misclassification is important for practical applications, including many data mining applications, such as targeted marketing, fraud and intrusion detection, among others. Classification is often idealized as a problem where every example is equally important, and the cost of misclassification is always the same. The real world is messier. Typically, some examples are much more important than others, and the cost of misclassifying in one way differs from the cost of misclassifying in another way. A body of work on this subject has become known as costsensitive learning, in the areas of machine learning and data mining.

[0005]
Research in costsensitive learning falls into three main categories. The first category is concerned with making particular classifier learners costsensitive, including methods specific for decision trees (see, for example., U. Knoll, G. Nakhaeizadeh, and B. Tausend, “Costsensitive pruning of decision trees”, Proceedings of the Eight European Conference on Machine Learning, pp. 383386, 1994, and J. Bradford, C. Kunz, R. Kohavi, C. Brunk, and C. Brodley, “Pruning decision trees with misclassification costs”, Proceedings of the European Conference on Machine Learning, pp. 131136, 1998), neural networks (see, for example, P. Geibel and F. Wysotzki, “Perceptron based learning with example dependent and noisy costs”, Proceedings of the Twentieth International Conference on Machine Learning, 2003), and support vector machines (see, for example, G. Fumera and F. Roli, “Costsensitive learning in support vector machines”, VIII Convegno Associazione Italiana per L'Intelligenza Artificiale, 2002). The second category uses Bayes risk theory to assign each example to its lowest expected cost class (see, for example, P. Domingos, “MetaCost: A general method for making classifiers cost sensitive”, Proceedings of the Fifth International Conference on Knowledge Discovery and Data Mining, pp. 144164, ACM Press, 1999, and D. Margineantu, Methods for CostSensitive Learning, PhD thesis, Department of Computer Science, Oregon State University, Corvallis, 2001). This requires classifiers to output class membership probabilities and sometimes requires estimating costs (see, B. Zadrozny and C. Elkan, “Learning and making decisions when costs and probabilities are both unknown”, Proceedings of the Seventh International Confernece on Knowledge Discovery and Data Mining, pp. 204213, ACM Press, 2001) (when the costs are unknown at classification time). The third category concerns methods that modify the distribution of training examples before applying the classifier learning method, so that the classifier learned from the modified distribution is costsensitive. We call this approach costsensitive learning by example weighting. Work in this area includes stratification methods (see, for example, P. Chan and S. Stolfo, “Toward scalable learning with nonuniform class and cost distributions”, Proceedings of the Fourth International Conference on Knowledge Discovery and Data Mining, pp. 164168, 1998, and L. Breiman, J. H. Friedman, R. A. Olsen, and C. J. Stone, Classification and Regression Trees, Wadsworth International Group, 1984) and the costing algorithm (see, for example, B. Zadrozny, J. Langford, and N. Abe, “Costsensitive learning by costproportionate example weighting”, Proceedings of the Third IEEE International Conference on Data Mining, pp. 435442, 2003). This approach is very general since it reuses arbitrary classifier learners and does not require accurate class probability estimates from the classifier. Empirically this approach attains similar or better costminimization performance.

[0006]
Unfortunately, current methods in this category suffer from a major limitation: they are wellunderstood only for twoclass problems. In the twoclass case, it is easy to show that each example should be weighted proportionally to the difference in cost between predicting correctly or incorrectly (see, again, Zadrozny et al., ibid.). However, in the multiclass case there is more than one way in which a classifier can make a mistake, breaking the application of this simple formula. Heuristics, such as weighting examples by the average misclassification cost, have been proposed (see, again, Breiman et al., ibid., and the Margineantu thesis, ibid.), but they are not wellmotivated theoretically and do not seem to work very well in practice when compared to methods that use Bayes risk minimization (see, again, Domingos, ibid.).
SUMMARY OF THE INVENTION

[0007]
It is therefore an object of the present invention to provide a method for multiclass costsensitive learning based on an example weighting scheme.

[0008]
According to the invention, the methods are based on example weighting schemes that are derived using two key ideas: 1) data space expansion and 2) gradient boosting with stochastic ensembles. The latter is a formal framework that give rise to a coherent body of methods.

[0009]
One of the methods of invention, which is based on the idea 1) above, works by repeatedly sampling from the expanded data set, which is obtained by enhancing each example in the original data set with as many data points as there are possible labels for any single instance. It then repeatedly draws subsample from this expanded data set using weighted sampling according to a certain example weighting scheme, in which each labeled example is given the weight specified as the difference between the maximum possible misclassification cost for the instance in question and the misclassification associated with the label in the particular labeled example. The example weighting remains constant throughout the iterative sampling procedure. It then finally outputs a classifier hypothesis which is the average of all the hypotheses output in the respective iterations.

[0010]
Another one of the methods of invention, which is based on the idea 2) above, works by iteratively applying weighted sampling from the same expanded data set, using a different weighting scheme. The weighting scheme of this method gives each labeled example the weight specified as the difference between the average cost on that instance by the averaged hypotheses from the iterations so far and the misclassification cost associated with the label in the labeled example in question. Emphatically, the weighting changes in every iteration, since it depends on the performance of the averaged hypothesis obtained up to the current iteration. Additionally, the example weights used in this method can be both positive and negative, since the label given in any labeled example does not necessarily correspond to the best label for the given instance, i.e. the label with the minimum cost, due to the use of data space expansion. Negative weights do not admit the use of weighted sampling. The method deals with this problem by calling the component classification algorithm on a modified binary classification problem in which each example is itself already a labeled pair, and its (meta) label is 1 or 0 depending on whether the example weight in the above weighting scheme is positive or negative, respectively.

[0011]
The results of the methods of invention are obtained by outputting all of the classifier representations obtained through the iterations, and represent the average over them. These representations can be arbitrary representations of classifiers, such as decision trees, neural networks and support vector machines, for the problem at hand, such as network intrusion detection, fraud detection, targeted marketing, credit risk rating, among other things. For example, in the application to network intrusion detection, each one of these representations could be a decision tree that specifies a set of conditions on various attributes of a network connection event, which together signal certain types of network intrusion. Such representations can be further applied on a new network connection to output judgment whether or not the connection is to be suspected to be some type of an intrusion attempt with reasonable likelihood, and decisions can be based on this judgment to determine the appropriate course of action, such as denial of service or probing.
BRIEF DESCRIPTION OF THE DRAWINGS

[0012]
The foregoing and other objects, aspects and advantages will be better understood from the following detailed description of a preferred embodiment of the invention with reference to the drawings, in which:

[0013]
FIG. 1 is a block diagram showing the architecture of the system implementing one of the methods according to the invention;

[0014]
FIG. 2 is a flow chart showing the logic of the method for multiclass costsensitive learning implemented on the system shown in FIG. 1;

[0015]
FIG. 3 is a block diagram showing the architecture of the system implementing another one of the methods according to the invention;

[0016]
FIG. 4 is a flow chart showing the logic of the method for multiclass costsensitive learning implemented on the system shown in FIG. 3; and

[0017]
FIG. 5 is an example of a decision tree to illustrate the process implemented by the invention.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION

[0018]
We begin by introducing some general concepts and notation we use in the rest of the description.
CostSensitive Learning and Related Problems

[0019]
A popular formulation of the costsensitive learning problem is via the use of a cost matrix. A cost matrix, C(y_{1}, y_{2}), specifies how much cost is incurred when misclassifying an example labeled y_{2 }as y_{1}, and the goal of a costsensitive learning method is to minimize the expected cost. Zadrozny and Elkan (B. Zadrozny and C. Elkan, “Learning and making decisions when costs and probabilities are both unknown”, Proceedings of the seventh International Conference on Knowledge Discovery and Data Mining, pp. 204213, ACM Press, 2001) noted that this formulation is not applicable in situations in which misclassification costs depend on particular instances, and proposed a more general form of cost function, C(x, y_{1 }, y_{2}), that allows dependence on the instance x. Here we adopt this general formulation, but note that in the reasonable case in which the cost is minimized by the true label, we can drop the redundant information y_{2 }and write C(x, y_{1}) for C(x, y_{1}, y_{2}).

[0020]
Once we allow the costs to depend on each example, it is natural to assume that the costs are generated according to some distribution, along with the examples, which leads to the following formulation. In (multiclass) cost sensitive classification, examples of the form (x, (C(x, y
_{1}), . . . , C(x, y
_{k})) are drawn from a distribution D over a domain X×R
^{+} ^{ k }. (Throughout, we will let k denote Y.) Given a set of examples, S=(x
_{i}, (C(x
_{i}, y))
_{yεY})
^{m}, the goal is to find a classifier h:X→{1, . . . , k} which minimizes the expected cost of the classifier:
$\begin{array}{cc}\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{min}}{E}_{D}\left[C\left(x,h\left(x\right)\right)\right]& \left(1\right)\end{array}$
We can assume without loss of generality that the costs are normalized so that
$\forall x\in X\text{\hspace{1em}}\underset{x\in Y}{\mathrm{min}}C\left(x,y\right)=0.$
Note that with this normalization, the above formulation of cost is equivalent to the common fonnulation in terms of misclassification cost, i.e.,
$\underset{h}{\mathrm{min}}{E}_{D}\left[C\left(x,h\left(x\right)\right)I\left(h\left(x\right)\ne \mathrm{arg}\text{\hspace{1em}}\underset{y}{\mathrm{min}}C\left(x,y\right)\right)\right]$
Nonmally a learning method attempts to do this by minimizing the empirical cost in the given training data, given some hypothesis class
:
$\begin{array}{cc}\mathrm{arg}\text{\hspace{1em}}\underset{h\in H}{\mathrm{min}}\sum _{\left(x,{\langle C\left(x,y\right)\rangle}_{y\in Y}\right)\in S}C\left(x,h\left(x\right)\right)& \left(2\right)\end{array}$
We note that we sometimes use the empirical expectation notation, Ê, to refer to the averaged empirical cost, namely
${\hat{E}}_{\left(x,{\langle C\left(x,y\right)\rangle}_{y\in Y}\right)S}C\left(x,h\left(x\right)\right)=\frac{1}{\uf603S\uf604}\sum _{\left(x,{\langle C\left(x,y\right)\rangle}_{y\in Y}\right)\in S}C\left(x,h\left(x\right)\right)$
As a building block of our method, we make use of methods for solving importance weighted classification problems, which we define below. In importance weighted classification, examples of the form (x, y, c) are drawn from a distribution D over a domain X×Y×R
^{+}. Given a set of examples S=(x, y, c)
^{m}, the goal is to find a classifier h:X→Y having minimum importanceweighted misclassificaton error:
$\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{min}}{E}_{\left(x,y,c\right)D}c\xb7I\left(h\left(x\right)\ne y\right)$
Again, usually, a learning method attempts to meet this goal by minimizing the empirical weighted error in some hypothesis class
:
$\begin{array}{cc}\mathrm{arg}\text{\hspace{1em}}\underset{h\in H}{\mathrm{min}}\sum _{\left(x,y,c\right)\in S}c\xb7I\left(h\left(x\right)\ne y\right)& \left(3\right)\end{array}$
We note that importance weighted classification can be solved very well with a classification method, by use of weighted rejection sampling techniques (see,again, Zadronzny, Langford, and Abe, ibid.).
Hypothesis Representations and Other Notation

[0021]
In the above, we assumed that the hypotheses output by a costsensitive learner is a functional hypothesis h, i.e., h:X→Y. It is also possible to allow hypotheses that are stochastic, namely
h:X×Y→[0,1]
subject to the stochastic condition:
$\forall x\in X\sum _{y\in Y}h\left(x,y\right)=1.$
With stochastic hypotheses, stochastic costsensitive learning is defined as that of minimizing the following expected cost:
$\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{min}}{E}_{D}\sum _{y\in Y}C\left(x,y\right)h\left(x,y\right)$
Note that in the special case that h is deterministic, this formulation is equivalent to the definition given in Equation (1). Also, this is a convexification of the standard objective function that we usually expect a stochastic costsensitive learner to minimize, i.e.,
${E}_{D}\left[C\left(x,\mathrm{arg}\text{\hspace{1em}}\underset{y\in Y}{\mathrm{max}}h\left(x,y\right)\right)\right]$
We also consider a variant of costsensitive learning in which relational hypotheses are allowed. Here relational hypotheses h are relations over X×Y, i.e., h:X×Y→{0, 1}. In general h is neither functional nor stochastic, and in particular it may violate the stochastic condition,
$\sum _{y\in Y}h\left(x,y\right)=1.$

[0022]
We often use the more general notation of h(x,y), meant for stochastic and relational hypotheses, even when h is a deterministic function from X to Y. As notational shorthand, for a stochastic hypothesis h, we write h(x) to denote h(x, ·):Y→[0,1], and C(x, h(x)) to denote the expected cost of its predictions, i.e.,
$C\left(x,h\left(x\right)\right)=\sum _{y\in Y}h\left(x,y\right)C\left(x,y\right).$
Finally, we note that we often write “x ε S” as a shorthand for “
y ε Y(x,y) ε S”.
The Methodology

[0023]
Our methodology can be interpreted as a reduction, which translates a multiclass costsensitive learning problem to a classifier learning problem. That is, it allows us to solve the costsensitive learning problem using an arbitrary classifier learning method as a component algorithm. This methodology is derived using two key ideas: 1) expanding data space and 2) gradient boosting with stochastic ensembles. Theoretical performance guarantee on a particular variant of the invented methodology is derived using a convexification of the objective function by the expected cost function. Below we will explain these two key ideas by exhibiting a prototypical method based on each.

[0024]
A representative method in the prior art of iterative methods for costsensitive learning is the method proposed in Zadrozny, Langford and Abe, ibid., called costing. The weighting scheme of this method exploits the following observation: For the binary class case, the above formulation in terms of cost per example, C(x, y_{2}), can be further reduced to a formulation in terms of a single importance number per example. This is possible by associating a number indicating the importance of an example (x, y_{2}), given by C(x, 0)−C(x, 1). This conversion allows us to reduce the costsensitive learning problem to a weighted classifier learning problem, but it has not been known how that would be done for the multiclass scenario. It is therefore natural to consider iterative weighting schemes, in which example weights are iteratively modified in search for the optimal weighting.

[0025]
A straightforward application of iterative weighting suffers from an inability to directly take into account the different costs associated with multiple ways of misclassifying examples. This translates to nonconvergence of the method in practice. We address this issue by the technique of expanding data space, the first of the two key ideas.

[0000]
Data Space Expansion

[0026]
The objective of minimizing the empirical cost on the original training sample is equivalent to minimization on the following expanded sample. Given a labeled sample S consisting of (x, (C(x,y))_{yεY}) of size m, we define a sample S′ of size mk for classification, where k is the size of the label set, i.e., k=Y, as follows.
${S}^{\prime}=\left\{\left(x,y\right),\underset{x,y}{\mathrm{max}}C\left(x,y\right)C\left(x,y\right)\right)\u2758\left(x,{\langle C\left(x,y\right)\rangle}_{y\in Y}\right)\in S,y\in Y\}$
Minimizing the importance weighted loss,
$\sum _{\left(x,y,c\right)\in {S}^{\prime}}c\xb7I\left(h\left(x\right)\ne y\right)$
on this new dataset also minimizes the cost on our original sample. The algorithm DSE (Data Space Expansion) takes advantage of this observation, which is summarized below as a theorem.

[0027]
THEOREM 1. With the definitions given in FIG. 3, a hypothesis IL minimizing the weighted classification error on the expanded weighted sample S′,
$\underset{\left(x,y,c\right)~{S}^{\prime}}{\hat{E}}\left[c\xb7I\left(h\left(x\right)\ne y\right)\right]$
also minimizes the cost on the original sample S,
$\underset{\left(x,y,c\right)~{S}^{\prime}}{\hat{E}}\left[{C}_{h\left(x\right)}\right].\text{}\begin{array}{c}\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{min}}\underset{\left(x,y,c\right)~{S}^{\prime}}{\hat{E}}\left[c\xb7I\left(h\left(x\right)\ne y\right)\right]\\ =\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{min}}\underset{\left(x,\stackrel{~}{C}\right)~S}{\hat{E}}\sum _{y\in Y}\left[\left(\underset{{y}^{\prime}\in Y}{\mathrm{max}}{C}_{{y}^{\prime}}{C}_{y}\right)\xb7I\left(h\left(x\right)\ne y\right)\right]\\ =\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{max}}\underset{\left(x,\stackrel{~}{C}\right)~S}{\hat{E}}\sum _{y\in Y}\left[{C}_{y}\xb7I\left(h\left(x\right)\ne y\right)\right]\\ =\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{max}}\underset{\left(x,\stackrel{~}{C}\right)~S}{\hat{E}}\left[\left(\sum _{y\in Y}{C}_{y}\right){C}_{h\left(x\right)}\right]\\ =\mathrm{arg}\text{\hspace{1em}}\underset{h}{\mathrm{max}}\underset{\left(x,\stackrel{~}{C}\right)~S}{\hat{E}}\left[{C}_{h\left(x\right)}\right]\end{array}$
Gradient Boosting with Stochastic Ensembles

[0028]
Having described the idea of data space expansion, we now cmbine it with the gradient boosting framework to arrive at our main method. In particular, we cast the stochastic multiclass costsensitive learning in the framework of gradient boosting (see L. Mason, J. Baxter, P. Barlett, and M. Frean, “Boosting algorithms as gradient descent”, Advances in Neural Information Processing Systems 12, pp. 512518, 2000), with the objective function defined as the expected cost of the stochastic ensemble, obtained as a mixture of individual hypotheses, on the expanded data set. As we stated above, a functional hypothesis of the form h:X→Y can be viewed as a special case of a stochastic hypothesis. We then define a stochastic ensemble hypothesis H, given multiple functional hypotheses, h_{t}, t=1, . . . , T, as the conditional distribution defined as the mixture of the component hypotheses, namely,
$\forall x\in X,\forall y\in Y,H\left(x,y\right)=\sum _{t=1}^{T}{h}_{t}\left(x,y\right)$
Let H_{t }denote the mixture hypothesis of the learning procedure at round t. The procedure is to update its current combined hypothesis by the mixture of the previous combined hypothesis and a new hypothesis, i.e., by setting
H _{t}(x,y)=(1−β)H _{t1}(x,y)+βh(x,y)
Thus, the expected cost of H_{t }on x is
C(x, H _{t}(x))=(1−β)C(x, H _{t1}(x))+βC(x, h _{t}(x))
Now, suppose that h predicts a particular label y for x, i.e., h(x,y)=1, then
C(x, H _{t}(x))=(1−β)C(x, H _{t1}(x))+βC(x,y)
If we now take a derivative of this function with respect to β, we get
$\frac{\partial C\left(x,{H}_{t}\left(x\right)\right)}{\partial \beta}=C\left(x,y\right)C\left(x,{H}_{t1}\left(x\right)\right)$
Note that this is the difference between the average cost of the current ensemble hypothesis and the new weak hypothesis assigning probability one to the specified label.

[0029]
We then take this derivative with respect to all data points (x,y) in the expanded data set S′, and thus the gradient is mkdimensional. We then expect the weak learner to find a hypothesis h whose innerproduct with the negative gradient is large. That is, the output h of the weak learner seeks to maximize the following sum.
$\begin{array}{cc}\langle h,\nabla C\rangle =\frac{1}{W}\sum _{x\in S}\sum _{y\in Y}\left(C\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)\right)h\left(x,y\right)& \left(9\right)\end{array}$
where W denotes the sum of absolute values of the weights, i.e.,
$W=\sum _{x\in S}\sum _{y\in Y}\uf603C\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)\uf604.$
Note that unlike the weights typically used in existing hosting methods, the weights w_{x,y}:=C(x, H_{t1}(x))−C(x,y) can be negative, since y is not necessarily the best (least cost) label. This means that the weak learner now receives both positive and negative weights. While the minimization of weighted misclassification with positive and negative weights makes perfect sense as an optimization problem, its interpretation as a classification problem is not immediately clear. In particular, it prohibits the use of weighted sampling as a means of realizing the weighted classification problem.

[0030]
We deal with this problem by converting a relational version of the weighted multiclass classification problem (i.e., of finding h to maximize Equation 9) in each iteration to a weighted binary classification problem. Specifically, we convert each example pair (x,y) to ((x,y), l), and set l=1 if the weight on (x,y) is positive, and l=0 if the weight is negative. The output hypothesis of the binary classifier is in general relational, so it is converted to a stochastic hypothesis by the procedure Stochastic. (The particular way this procedure is defined is motivated by the theoretical guarantee, which will be shown in the next subsection.) The overall process, consisting of multiple iterations of such a reduction, constitutes a reduction of the stochastic multiclass costsensitive classification to binary weighted classification.

[0031]
With the foregoing definitions, we can now state our main method, GBSE (Gradient Boosting with Stochastic Ensembles).
Theoretical Performance Guarantee on a Variant

[0032]
It turns out that a strong theoretical performance guarantee can be proved on a variant of this method, which we describe below. We define the per label average cost, {tilde over (C)}(x, H(x)), of a stochastic hypothesis H, in general, as follows.
$\stackrel{~}{C}\left(x,H\left(x\right)\right)=\frac{1}{k}\sum _{y\in Y}H\left(x,y\right)C\left(x,y\right)$
Note that, with this definition, the empirical loss (cost) of H on the original sample S, C(H, S), can be expressed as the sum of this per label cost over the expanded data set S′={(x,y)x ε S, y ε Y}.
$\begin{array}{c}C\left(H,S\right)=\sum _{x}\sum _{y}H\left(x,y\right)C\left(x,y\right)\\ =\sum _{x}\sum _{y}\stackrel{~}{C}\left(x,H\left(x\right)\right)\end{array}$
The variant, for which we prove our theoretical performance guarantee is obtained by simply replacing the weight updating rule of GBSE by the following:
w _{x,y} ={tilde over (C)}(x, H _{t1}(x))−C(x,y)
The resulting variant, which we call GBSET (Gradient Boosting with Stochastic Ensembles—Theoretical version), is summarized in FIG. 5.

[0033]
We can show that GBSET has a boosting property given a version of weak learning condition on the component classifier. This weak learning condition, which we make precise below, is one that is sensitive to class imbalance.

[0034]
DEFINITION 1. We say that an algorithm A for the binary importance weighted classification problem, as defined above, satisfies the weak learning condition for a given classification sample S=(x,y)^{m}, if for arbitrary distribution over S, (w)^{m}, Σw=1, when it is given S′=(x,y,w)^{m }as input, its output h satisfies the following, for some fixed γ>0:
$\begin{array}{cc}\sum _{\left(x,y,w\right)\in {S}^{\prime}}w\xb7I\left(h\left(x\right)=y\right)\ge \sum _{y=0}w+\gamma \sum _{y=1}w& \left(12\right)\end{array}$

[0035]
THEOREM 2. Suppose that the component leaner A satisfies the weak learning condition for the input sample S. Then, the output of GBSET will converge to a stochastic ensemble hypothesis achieving minimum expected cost on the (original) sample S. In particular, if we set αt=α for all t,
$\sum _{x}\sum _{y}{H}_{T}\left(x,y\right)C\left(x,y\right)\le \mathrm{exp}\left\{\frac{\mathrm{\gamma \alpha}}{k}T\right\}\sum _{x}\sum _{y}{H}_{0}\left(x,y\right)C\left(x,y\right)$
Proof

[0036]
We first establish the following simple correspondence between the weak learning conditions on the relational multiclass classification problem that we wish to solve in each iteration, and the binary classification problem that is given to the component algorithm to solve it.

[0037]
DEFINITION 2. Let S be a weighted sample of the form S=(x,y,w)
^{m}, where weights w can be both positive and negative. Then define a transformed sample S′ from S by S′=((x,y), l, w)
^{m }where l=I(w≧0).

 1. The relational weighted multiclass classification problem for S is to find a relational hypothesis h:X×Y→{0, 1} that maximizes the following sum:
$a\left(h,S\right)=\frac{1}{W}\sum _{\left(x,y,w\right)\in S}w\xb7h\left(x,y\right)\text{\hspace{1em}}\mathrm{where}$
$W=\sum _{\left(x,y,w\right)\in S}\uf603w\uf604.$
 2. The weighted binary classification problem for S′ is to find a hypothesis h′:X×Y→{0, 1} that maximizes the following weighted classification accuracy:
${a}^{\prime}\left({h}^{\prime},{S}^{\prime}\right)=\frac{1}{W}\sum _{\left(\left(x,y\right),l,\uf603w\uf604\right)\in {S}^{\prime}}\uf603w\uf604\xb7I\left({h}^{\prime}\left(x,y\right)=l\right)$

[0040]
LEMMA 1. Assume the notation of Definition 2. Then, for arbitrary ε>0, h satisfies the following condition on the relational multiclass classification problem for S:
a(h,S)≧ε
if and only if (the same) h satisfies the corresponding condition on the transformed binary classification problem for S′:
${a}^{\prime}\left(h,{S}^{\prime}\right)\ge \frac{\sum _{l=0}\uf603w\uf604}{W}+\epsilon $
Proof of Lemma 1
$\begin{array}{c}W\xb7a\left(h,{S}^{\prime}\right)=\sum _{\left(\left(x,y\right),l,\uf603w\uf604\right)\in {S}^{\prime}}\uf603w\uf604\xb7I\left(h\left(x,y\right)=l\right)\\ =\sum _{w\ge 0}w\xb7I\left(h\left(x,y\right)=1\right)+\sum _{w<0}w\xb7I\left(h\left(x,y\right)=0\right)\\ =\sum _{w\ge 0}w\xb7h\left(x,y\right)+\sum _{w<0}w\left(1h\left(x,y\right)\right)\\ =\sum _{\left(x,y,w\right)\in S}w\xb7h\left(x,y\right)+\sum _{w<0}\uf603w\uf604\\ =W\xb7a\left(h,S\right)+\sum _{\left(x,y,w\right)\in S:w<0}\uf603w\uf604\end{array}$
Hence the lemma follows.
Proof of Theorem 2

[0041]
First, note that applying Stochastic to h, can increase the expected cost only for x's such that [yh_{t}(x,y)=1}=0, and for such x's the cost of f, equals that of H{t1} by the definition of Stochastic. Hence, the empirical cost of f, on the original sample S, C(f_{t}, S), satisfies the following:
$\begin{array}{cc}C\left({f}_{t},S\right)C\left({h}_{t},S\right)\le \sum _{x:\forall y\text{\hspace{1em}}h\left(x,y\right)=0}\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)& \left(13\right)\end{array}$
Now recall that the expected empirical cost of H_{t }equals the following, where we drop the subscript t from α_{t}.
$\begin{array}{cc}\begin{array}{c}C\left({H}_{t},S\right)=\sum _{x,y}\left(1\alpha \right){H}_{t1}\left(x,y\right)C\left(x,y\right)+\alpha \sum _{x,y}f\left(x,y\right)C\left(x,y\right)\\ =\sum _{x,y}\left(1\alpha \right)\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)+\alpha \sum _{x,y}f\left(x,y\right)C\left(x,y\right)\end{array}& \left(14\right)\end{array}$
Hence, by combining Equation 13 and Equation 14, we can show the following bound on the decrease in empirical cost in each iteration:
$C\left({H}_{t1},S\right)C\left({H}_{t},S\right)$
$\begin{array}{c}\text{\hspace{1em}}=\sum _{x}\alpha \left(\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\sum _{y}f\left(x,y\right)C\left(x,y\right)\right)\\ =\sum _{x}\alpha \left(\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\sum _{y}h\left(x,y\right)C\left(x,y\right)\right)+\\ \sum _{x}\alpha \left(\sum _{y}h\left(x,y\right)C\left(x,y\right)\sum _{y}f\left(x,y\right)C\left(x,y\right)\right)\ge \\ \sum _{x}\alpha \left(\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\sum _{y}h\left(x,y\right)C\left(x,y\right)\right)\\ \alpha \left(\sum _{x:\forall y\text{\hspace{1em}}h\left(x,y\right)=0}\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\right)\ge \\ \alpha (\sum _{x}(\sum _{y:h\left(x,y\right)=1}h\left(x,y\right)\left(\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)\right)+\\ \sum _{y:h\left(x,y\right)=0}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\sum _{x:\forall y\text{\hspace{1em}}h\left(x,y\right)=0}\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\\ =\alpha (\sum _{x}\left(\sum _{y}h\left(x,y\right)\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)\right)+\\ \left(\sum _{x}\sum _{y:h\left(x,y\right)=0}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\sum _{x:\forall y\text{\hspace{1em}}h\left(x,y\right)=0}\sum _{y}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)\right))\ge \\ \alpha \sum _{x}\sum _{y}h\left(x,y\right)\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right))\ge \\ \alpha \underset{x}{\gamma \sum}\sum _{y:\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)>0}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)C\left(x,y\right)\ge \\ \underset{x}{\mathrm{\alpha \gamma}\sum}\stackrel{~}{C}\left(x,{H}_{t1}\left(x\right)\right)=\frac{\mathrm{\alpha \gamma}}{k}C\left({H}_{t1},S\right)\end{array}$
In the above derivation, the second to last inequality follows from the weak learning condition and applying Lemma 1 with weights {tilde over (C)}(x, H_{t1}(x))−C(x,y). The last inequality follows from the fact that the weights are nonnalized so that the minimum achievable cost is zero for all x. Noting that the sum of these weights is positive whenever the current ensemble hypothesis is suboptimal, this guarantees a positive progress in each iteration unless optimality is achieved. Since the expected empirical cost function as defined by Σ_{x}Σ_{y }F(x,y) C(x,y) is convex (in fact linear), this implies convergence to the global optimum. Noting that in each iteration, the empirical cost is reduced at least by a factor of
$1\frac{\mathrm{\gamma \alpha}}{k},$
and the theorem follows.

[0042]
Note that at earlier iterations, the binary classifier used as the component learner is likely to be given weighted sample with balanced positive and negative examples. As the number of iterations increases and progress is made, however, it will receive samples that are increasingly more negative. (This is because the positive examples correspond to labels that can further improve the current performance.) It therefore becomes easier to attain high weighted accuracy by simply classifying all examples to be negative. The weak learning condition of Equation 12 appropriately deals with this issue, as it requires that the weak learner achieve better weighted accuracy than that attainable by assigning all examples to the negative class.
Variations

[0043]
In addition to the two variants of the Gradient Boosting with StochastiEnsembles method presented above, namely GBSE and GBSET, other related variations are possible. For example, in one variant, the weighted sampling can be done in two steps; the instance is sampled in the first step according to a probability proportional to
max_{y}×w_{x,y }
and then choosing the label y with a probability proportional to
w_{x,y}
In a yet another variant, the weighted sampling can be done in two steps; the instance is sampled in the first step according to the same probability as above, and for the chosen instance, examples are deterministically added for all possible labels.
Implementation

[0044]
Referring now to FIG. 1, there is shown a system on which a method for multiclass, costsensitive learning according to the invention may be implemented. This system comprises a multiclass costsensitive learning top control module 1 which controls the overall control flow, making use of various subcomponents of the system. A learning algorithm storage module 2 stores a representation of an algorithm for classification learning. An arbitrary algorithm for classification can be used here. Alternatively the learning algorithm can be a decision tree learning algorithm, a naïve Bayes method, a logistic regression method or neural networks. The model output module 3 stores the models obtained as a result of applying the learning algorithm stored in module 2 to training data given by weighted sampling module 4 and outputs a final model by aggregating these models. The weighted sampling module 4 accesses the data stored in data storage module 7, samples a relatively small subset of the data with acceptance probability determined using the example weights, and passes the obtained subsample to module 1. The weight update module 5 updates the example weights for sampling using a particular function determined by the current weights and current models. The model update module 6 updates the current model using the model's output in the previous iterations stored in the current model storage module 8 and the output model of the current iteration output by module 3 and stores the resulting updated model in module 8.

[0045]
FIG. 2 shows a flow diagram of the process implemented in the system of FIG. 1. The first three steps initialize the process. In Step 21, expanded data T is initialized using the input data S. In Step 22, H_{0 }is initialized by setting for all (x,y) in T. Finally, in Step 23, the weights for all (x,y) in Tare initialized. The iteration begins in the decision block of Step 4. A test is made to determine if i=t. If not, Step 25 performs the computation for all (x,y)
$w\left(x,y\right)=\left(\sum _{y\in Y}{H}_{t1}\left(x,y\right)\xb7C\left(c,y\right)\right)C\left(x,y\right)$
The decision block in Step 26 determines if there is more data in T or a STOP condition has been met. If not, in Step 27, (x,y) is sampled from T and accepted with a probability proportional to w(x,y). Next, in Step 28, if accepted, ((x,y), (w(x,y)>0)) is added to subsample T^{t}. A return is then made to the decision block in Step 26. When there is no more data in T or a STOP condition has been met, the process goes to Step 29 where the learning algorithm is run on T^{t }to obtain model h_{t}. Next, in Step 30, f_{t }is set equal to stoch(h_{t}). Then, in Step 31, α_{t }is chosen and H_{t }is set equal to (1−α_{t})H_{t1}+α_{t}f_{t}. The index i is incremented at Step 31, and a return is then made to the decision block in Step 24. If i=t, then in Step 33 the final model H_{t }is output.

[0046]
FIG. 3 shows a system on which another method for multiclass, costsensitive learning according to the invention may be implemented. This system is similar to that shown in FIG. 1 and comprises a multiclass costsensitive learning top control module 1 which controls the overall control flow, making use of various subcomponents of the system, a learning algorithm storage module 2, which stores a representation of an algorithm for classification learning, a model output module 3, which stores the models obtained as a result of applying the learning algorithm stored in module 2 to training data given by weighted sampling module 4 and outputs a final model by aggregating these models, and a weighted sampling module 4, which accesses the data stored in data storage module 7, samples a relatively small subset of the data with acceptance probability determined using the example weights, and passes the obtained subsample to module 1. The weight calculation module 5′ replaces the weight upadate module 5, which updates the example weights for sampling using a dynamically changing weighting scheme. The model update module 6 updates the current model using the model's output in the previous iterations stored in the current model storage module 8 and the output model of the current iteration output by module 3 and stores the resulting updated model in module 8.

[0047]
FIG. 4 shows a flow diagram of the process implemented in the system of FIG. 3. The first step initializes the process. In Step 41, expanded data T is initialized using the input data S. In Step 42, the weights for all (x,y) in Tare set. The iteration begins in the decision block of Step 43. A test is made to determine if i=t. If not, a test is made in Step 44 to determine if there is no more data in T or a stop condition has been met. If not, Step 45 samples (x,y) from T and accepts (x,y) with probability proportional to w(x,y). If accepted, (x,y) is added to subsample T′ in Step 46. The process then loops back to decision block in Step 44 until there is either no more data in T or a stop condition has been met. At this point, the learning algorithm is run in Step 47 on T′ to obtain a model h_{t}. In Step 48, α is chosen so that when i=1, α=0 and H_{t}=(1−α_{t})H_{t1}+α_{t}f_{t}. The index i is incremented at Step 49, and a return is then made to the decision block in Step 43. If i=t, then in Step 50 the final model H_{t }is output.

[0048]
As a concrete example of applying the method of the invention to a real world problem, we describe an application to network intrusion detection. Network intrusion detection has recently become a prototypical application problem for multiclass, costsensitive learning. The multiclass aspect is essential because in this application there are typically more than one level of intrusion detection, such as probing and denial of service. The costsensitive aspect is important because vastly different costs are associated with different types of misclassification (e.g., false negatives are usually a magnitude more costly than false positives) and it is absolutely critical that any learning method used to derive an intrusion detection rule is sensitive to this cost structure.

[0049]
A network intrusion detection system based on the method and system of the invention for multiclass, costsensitive learning consists of the following steps:

 1) Convert past network connection data to a set of feature vectors, by mapping information on a network connection to a feature vector.
 2) Label each of these vectors with known labels, such as “normal”, “probe”, “denial of service”, or specific types of intrusions.
 3) Apply the method of the invention on the above data set, and obtain a classification rule.
 4) Convert new network connection data to feature vectors, apply the above classification rule to them, and flag those connections corresponding to feature vectors that are classified as different types of “intrusions” as such.

[0054]
A typical set of features used to transform connection data into a welldefined feature vector is that used in the network intrusion data set known as “KDD CUP 99” data, which is publically available. Here is the list of features in this data set (given in three separate tables).


Basic Features of Individual TCP Connections 
feature name  description  type 

duration  length (number of seconds) of the  continuous 
 connection 
protocol_type  type of protocol, e.g., TCP, UDP, etc.  discrete 
service  network service on the destination, e.g.,  discrete 
 http, telnet, etc. 
src_bytes  number of data bytes from source to  continuous 
 desitination 
dst_bytes  number of data bytes from destination to  continuous 
 source 
flag  normal or error status of the connection  discrete 
land  1 if connection is from/to the same  discrete 
 host/port; 0 otherwise 
wrong_fragment  number of “wrong” fragments  continuous 
urgent  number of urgent packets  continuous 


[0055]


Content Features Within a Connection 
Suggested by Domain Knowledge 
feature name 
description 
type 

hot 
number of “hot” indicators 
continuous 
num_failed_logins 
number of failed login attempts 
continuous 
logged_in 
1 if successfully logged in; 0 
discrete 

otherwise 
num_compromised 
number of “compromised” conditions 
continuous 
root_shell 
1 if root shell is obtained; 0 
discrete 

otherwise 
su_attempted 
1 if “suroot” command attempted; 0 
discrete 

otherwise 
num_root 
number of “root” accesses 
continuous 
num_file_creations 
number of file creation operations 
continuous 
num_shells 
number of shell prompts 
continuous 
num_access_files 
number of operations on access 
continuous 

control files 
num_outbound_cmds 
number of outbound commands in an 
continuous 

ftp session 
is_hot_login 
1 if the login belongs to the “hot” 
discrete 

list; 0 otherwise 
is_guest_login 
1 if the login is a “guest” login; 0 
discrete 

otherwise 


[0056]


Traffic Features Computed Using 
a TwoSecond Time Window 
feature name 
description 
type 

count 
number of connections to the same host 
continuous 

as the current connection in the past two 

seconds 

Note: The following features refer to 

these same host connections. 
serror_rate 
% of connections that have “SYN” 
continuous 

errors 
rerror_rate 
% of connections that have “REJ” 
continuous 

errors 
same_srv_rate 
% of connections of the same service 
continuous 
diff_srv_rate 
% of connections of different services 
continuous 
srv_count 
number of connections to the same 
continuous 

service as the current connection in the 

past two seconds 

Note: The following features refer to 

these sameservice connections. 
srv_server_rate 
% of connections that have “SYN” 
continuous 

errors 
srv_rerror_rate 
% of connections that have “REJ” 
continuous 

errors 
srv_diff_host_rate 
% of connections to different hosts 
continuous 


[0057]
As a result of applying the multiclass, costsensitive learning method of the invention to a data set consisting of these features and the corresponding labels, using a decision tree algorithm as the “classification learning algorithm” stored in Module 2 of FIG. 1, one obtains, as the classification rule, a voting function over a number of decision trees, such as the tree shown in FIG. 5.

[0058]
The system diagram of FIG. 1 and the flow chart of FIG. 2 illustrate a preferred embodiment of the invention, which corresponds to the method “GBSE” described herein. However, it will be understood by those skilled in the art that the method “DSE”, also described herein, may be used in the alternative. The main difference between DSE and GBSE is that in DSE, the sampling weights remain unchanged throughout all iterations. Consequently, the modules and funcationalities that are related to weight updating are unnecessary.
Experimental Evaluation

[0059]
We use the C4.5 decision tree learner described by J. Quinlan in C4.5: Programs for Machine Learning, Morgan Kaufmann (1993), as the base classifier learning method, because it is a standard for empirical comparisons and it was used as the base learner by Domingos for the MetaCost method (see, P. Domingos, “MetaCost: A general method for making classifiers cost sensitive”, Proceedings of the Fifth International Conference on Knowledge Discovery and Data Mining, pp. 155164, ACM Press, 1999).

[0060]
We compare our methods against three representative methods: Bagging (see L. Breiman, “Bagging predictors”, Machine Learning, 24(2):123140, 1996), Averaging cost (see, P. Chan and S. Stolfo, “Toward scalable learning with nonuniform class and cost distributions”,
Proceedings of the Fourth International Conference on Knowledge Discovery and Data Mining, pp. 164168, 1998), and MetaCost (see, Domingos, ibid.). The Averaging cost method was also used for comparison in Domingos, ibid. Note that Bagging is a costinsensitive learning method. Here we give a brief description of these methods, and refer the reader to Breiman, ibid., and Domingos, ibid., for the details.

 Bagging obtains multiple subsamples by sampling with replacement, feeds them to the base learner (C4.5), and takes the average over the ensemble of output hypotheses.
 Averaging Cost (AvgCost) obtains a subsample by weighted sampling with weights defined as the average cost for each x, and then feeds it to the base learner (C4.5).
 MetaCost uses bagging to obtain an ensemble of hypotheses, uses the ensemble to estimate the class probabilities, and then outputs a hypothesis that minimizes the expected risk with respect to these estimates.

[0064]
There are some deviations from these methods in our implementation, which we clarify below. The main deviation is that we use rejection sampling for all methods, while other sampling schemes such as resampling with replacement are used in the original methods. We do this for two reasons: (1) inadequacy of resampling with replacement, especially for C4.5, has been noted by various authors (see, for example, B. Zadrozny, J. Langford, and N. Abe, “Costsensitive learning by costproportionate example weighting”, Proceedings of the Third IEEE International Conference on Data Mining, pp. 435442, 2003); and (2) since our methods use rejection sampling, we do the same for the other methods for fairness of comparison. We stress that this deviation should only improve their performance. Another deviation is that we use a variant of MetaCost that skips the last step of learning a classifier on a relabeled training data set. It has been observed that this variant performs at least as well as MetaCost, in terms of cost minimization. (This variant has been called BagCost by D. Margineantu in Methods for CostSensitive Learning, PhD thesis, Department of Computer Science, Oregon State University, Corvallis, Oreg., 2001.) Also, in our implementation of AvgCost, we perform weighted sampling multiple times to obtain an emsemble of hypotheses, then output their average as the final hypothesis. We note that, due to our normalization assumption that the minimum cost for each instance x is always zero, our version of AvgCost is identical to a more sophisticated variant in which the difference between the average cost and the minimum cost is used for sampling weights. Our experience shows that this variant of AvgCost performs better than the original method.

[0065]
The methods were applied to five benchmark datasets available from the UCI machine learning repository (C. L. Blake and C. J. Merz, “UCI repository of machine learning databases”, Department of Information and Computer Sciences, University of California, Irvine, Calif., 1998) and one dataset from the UCI KDD archive (S. D. Bay, “UCI archive”, Department of Information and Computer Sciences, University of California, 2000). These datasets were selected by the criteria of having approximately 1,000 data or more, besides being multiclass problems. A summary of these datasets is given in Table 1.
TABLE 1 


Data set characteristics: data size, number of classes, 
and the ratio between the frequency of the most 
common class to the least common. 
 Dataset  # of examples  # of classes  Class ratio 
 
 Annealing  898  5  0.01316 
 KDD99  197710  5  0.0001278 
 Letter  20000  26  0.9028 
 Satellite  6435  6  0.4083 
 Solar flare  1389  7  0.002562 
 Splice  3190  3  0.4634 
 
Except for the KDD99 dataset, these datasets do not have standard misclassification costs associated with them. For this reason, we follow Domingos and generate cost matrices according to a model that gives higher costs for misclassifying a rare class as a frequent one, and inversely for lowest. (Note therefore that our experiments do not exploit the full generality of the instancedependent cost formulation presented above.) This reflects a situation that is found in many practical data mining applications, including direct marketing and fraud detection, where the rare classes are the most valuable to identify correctly.

[0066]
Our cost model is as follows: Let {circumflex over (P)}(y_{1}) and {circumflex over (P)}(y_{2}) be the empirical probabilities of occurrence of classes y_{1}and y_{2 }in the training data. We choose the nondiagonal entries of the cost matrix C(y_{1}, y_{2}), y_{1}≠y_{2 }with uniform probability from the interval [0,2000 {circumflex over (P)}(y_{1})/{circumflex over (P)}(y_{2})]. In Domingos, ibid., the diagonal entries were then chosen from the interval [0,1000], which often leads to cost matrices in which the correct label is not the least costly one. Besides being unreasonable (see C. Elkan, “Magical thinking in data mining: Lessons from coil challenge 2000”, Proceedings of the Seventh International Conference on Knowledge Discovery and Data Mining, pp. 42643 1, ACM Press, 1999), these cost matrices can give an unfair advantage to costsensitive methods over costinsensitive ones. We therefore set the diagonal entries to be identically zero, which is consistent with our normalization assumption.

[0067]
In all experiments, we randomly select ⅔ of the examples in the dataset for training and use the remaining ⅓ for testing. Also, for each training/test split we generate a different cost matrix according to the rules above. Thus, the standard deviations that we report reflect both variations in the data and in the misclassification costs.

[0068]
We remark on certain implementation details of the proposed learning methods in our experimentation. First, we note that in all of the methods used for comparison, C4.5 was used as the component algorithm, and the final hypothesis is expressed as an ensemble of output decision tress of C4.5. Its output hypothesis is therefore also an ensemble of decision trees. Next, the choice of the mixture weight α_{t }was unspecified in the algorithm descriptions. The choice of α_{t }was set at 1/t for most methods.

[0069]
The results of these experiments are summarized in Tables 2 and 3.
TABLE 2 


Experimental results: the average cost and standard error. 
Dataset  Bagging  AvgCost  MetaCost  DSE  GBSE 

Annealing  1059 ± 174  127.4 ± 12.2  206.8 ± 42.8  127.1 ± 14.9  33.72 ± 4.29 
Solar  5403 ± 397  237.8 ± 37.5  5317 ± 390  110.9 ± 28.7  48.17 ± 9.52 
KDD99  319.4 ± 42.2  42.43 ± 7.95  49.39 ± 9.34  46.68 ± 10.16  1.69 ± 0.78 
letter  151.0 ± 2.58  91.90 ± 1.36  129.6 ± 2.44  114.0 ± 1.43  84.63 ± 2.24 
Splice  64.19 ± 5.25  60.78 ± 3.65  49.95 ± 3.05  135.5 ± 14  57.50 ± 4.38 
Satellite  189.9 ± 9.57  107.8 ± 5.95  104.4 ± 6.43  116.8 ± 6.28  93.05 ± 5.57 


[0070]
TABLE 3 


Experimental results: the average data size used by each method 
in 30 iterations, and standard error. 
Dataset 
Bagging 
AvgCost 
MetaCost 
DSE 
GBSE 

Annealing 
11991 ± 13.1 
1002.8 ± 183 
11987 ± 9.84 
3795.5 ± 688 
1260.2 ± 224 
Solar 
18499 ± 20.4 
334.80 ± 37.5 
18510 ± 14.4 
2112.8 ± 276 
486.45 ± 53.3 
KDD99 
395310 ± 143 
2551.9 ± 428.6 
395580 ± 143 
12512 ± 2450 
4181 ± 783.6 
letter 
40037 ± 44.3 
159720 ± 2028 
40052 ± 41 
479130 ± 2710 
363001 ± 5557 
Splice 
42515 ± 26.6 
33658 ± 1697 
42501 ± 21 
52123 ± 592 
50284 ± 3659 
Satellite 
86136 ± 123 
60876 ± 1641 
85984 ± 127 
218870 ± 6516 
140810 ± 3335 


[0071]
Table 2 lists the average costs attained by each of these methods on the 6 data sets, and their stand errors. These results were obtained by averaging over 20 runs, each run consisting of 30 iterations of the respective learning method. These results appear quite convincing: GBSE outperforms all comparison methods on all data sets, except on Splice, for which it ranks second after MetaCost. Also, GBSE is the best performing among the proposed methods, confirming our claim that the combination of various techniques involved is indeed necessary to attain this level of performance.

[0072]
Table 3 lists the average total data size used by each of the methods in 30 iterations. Examining these results in conjunction with the data characteristics in Table 1 reveals a definite trend. First, note that the data sets are divided into to groups: those having very large skews, or very low class ratios (Annealing, KDD99 and Solar flare), and those having moderate skews (Satellite, Splice and Letter). It is evident that the methods based on example weighting (AvgCost, GBSE, DSE) use magnitudes smaller data sizes for the three data sets in the first group (i.e., with large skews), as compared to other methods, Bagging and MetaCost. The performance of GBSE is especially impressive on this group, achieving much lower costs while requiring very small data sizes. It is worth mentioning that it is these data sets in the first group with large skews that require costsensitive learning the most.

[0073]
We have provided a novel method for multiclass costsensitive learning based on gradient boosting with stochastic ensembles. It is not the first time that the issue of incorporating costsensitivity to boosting has been addressed. For example, AdaCost (see W. Fan, S. J. Stolfo, J. Zhang, and P. K. Chan, “AdaCost: Misclassification costsensitive boosting”, Proceedings of the Sixteenth International Conference on Machine Learning, pp. 97105, 1999) suggested a way of modifying AdaBoost's exponential loss using a function (called cost adjustment function) of the cost and confidence. The rational choice of this cost adjustment function, however, appears not to be wellunderstood. The stochastic ensemble that we employ in this method provides a straightforward but reasonable way of incorporating cost and confidence; i.e., in terms of expected cost.

[0074]
While the invention has been described in terms of a single preferred embodiment, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims.