Data Mining: Practical Machine Learning Tools and Techniques, Second Edition

248 CHAPTER 6| IMPLEMENTATIONS: REAL MACHINE LEARNING SCHEMES MakeModelTree (instances) { SD = sd(instances) for each k-valued n ...

6.5 NUMERIC PREDICTION 249 by split, and pruning it from the leaves upward, performed by prune. The node data structure contains ...

used to conflate these two nodes into the slightly more comprehensible tree that is shown. Rules from model trees Model trees ar ...

6.5 NUMERIC PREDICTION 251 in Section 3.3, and sometimes the structure can be expressed much more con- cisely using a set of rul ...

To use locally weighted regression, you need to decide on a distance-based weighting scheme for the training instances. A common ...

6.5 NUMERIC PREDICTION 253 nique. It also compares favorably with far more sophisticated ways of enhanc- ing Naïve Bayes by rela ...

weighted learning, primarily in the context of regression problems. Frank et al. (2003) evaluated the use of locally weighted le ...

6.6 CLUSTERING 255 independently. One way to split a cluster is to make a new seed, one standard deviation away from the cluster ...

The procedure is best illustrated by an example. We will use the familiar weather data again, but without the playattribute. To ...

6.6 CLUSTERING 257 the first five instances, there is no such host: it is better, in terms of category utility, to form a new le ...

an incremental way of restructuring the tree to compensate for incorrect choices caused by infelicitous ordering of examples. Th ...

6.6 CLUSTERING 259 Versicolor Versicolor Versicolor Virginica Versicolor Virginica Versicolor Versicolor Versicolor Versicolor V ...

experimentation with the acuity parameter was necessary to obtain such a nice division. The clusterings produced by this scheme ...

6.6 CLUSTERING 261 ing the values of attributes of instances in that cluster—that is, Pr[ai=vij|C] is a better estimate of the ...

(si) and for the data over all clusters (si), and use these in the category utility formula: Now the problem mentioned previous ...

6.6 CLUSTERING 263 distribution gives the probability that a particular instance would have a certain set of attribute values if ...

where f(x;mA,sA) is the normal distribution function for cluster A, that is: The denominator Pr[x] will disappear: we calculate ...

6.6 CLUSTERING 265 ization process makes the final result correct. Note that the final outcome is not a particular cluster but r ...

where the probabilities given the clusters A and B are determined from the normal distribution function f(x;m,s). This overall l ...

6.6 CLUSTERING 267 When the dataset is known in advance to contain correlated attributes, the independence assumption no longer ...