anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Technion</strong> - Computer Science Department - Ph.D. Thesis PHD-2008-12 - 2008<br />
Chapter 4<br />
Anytime Learning of<br />
Cost-sensitive Trees<br />
Assume, <strong>for</strong> example, that a medical center has decided to use machine <strong>learning</strong><br />
techniques to induce a diagnostic tool from records of previous patients. The<br />
center aims to obtain a comprehensible model, with low expected test costs and<br />
high expected accuracy. Moreover, in many cases there are costs associated with<br />
the predictive errors. In such a scenario, the task of the inducer is to produce a<br />
model with low expected test costs and low expected misclassification costs.<br />
In this chapter we build on the LSID3 algorithm, presented in Chapter 3, and<br />
introduce ACT (Anytime <strong>learning</strong> of Cost-sensitive Trees), our proposed <strong>anytime</strong><br />
framework <strong>for</strong> induction of cost-sensitive decision trees (Esmeir & Markovitch,<br />
2007a). ACT attempts to minimize the total cost of classification, that is the<br />
sum of testing costs and misclassification costs. ACT takes the same sampling<br />
approach as LSID3. The three major components of LSID3 that need to be<br />
replaced in order to adapt it <strong>for</strong> cost-sensitive problems are: (1) sampling the<br />
space of trees, (2) evaluating a tree, and (3) pruning a tree.<br />
4.1 Biasing the Sample Towards Low-cost Trees<br />
LISD3 uses SID3 to bias the samples towards small trees. In ACT, however,<br />
we would like to bias our sample towards low-cost trees. For this purpose, we<br />
designed a stochastic version of the EG2 algorithm, which attempts to build low<br />
cost trees greedily (Nunez, 1991). In EG2, a tree is built top-down, and the test<br />
that maximizes ICF is chosen <strong>for</strong> splitting a node, where,<br />
ICF (θ) = 2∆I(θ) − 1<br />
(cost (θ) + 1) w.<br />
71