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

in some other (previous or future) instance.TimeSeriesDeltareplaces attribute

values in the current instance with the difference between the current value and

the value in some other instance. In both cases instances in which the time-

shifted value is unknown may be removed, or missing values may be used.

Randomizing

Other attribute filters degrade the data.AddNoisetakes a nominal attribute and

changes a given percentage of its values. Missing values can be retained or

changed along with the rest.Obfuscateanonymizes data by renaming the rela-

tion, attribute names, and nominal and string attribute values.RandomProjec-

tionprojects the dataset on to a lower-dimensional subspace using a random

matrix with columns of unit length (Section 7.3). The class attribute is not

included in the projection.

Unsupervised instance filters

Weka’s instance filters, listed in Table 10.2, affect all instances in a dataset rather

than all values of a particular attribute or attributes.

Randomizing and subsampling

You can Randomizethe order of instances in the dataset.Normalizetreats all

numeric attributes (excluding the class) as a vector and normalizes it to a given

length. You can specify the vector length and the norm to be used.

There are various ways of generating subsets of the data. Use Resampleto

produce a random sample by sampling with replacement or RemoveFoldsto split

400 CHAPTER 10 | THE EXPLORER

Table 10.2 Unsupervised instance filters.

Name Function

NonSparseToSparse Convert all incoming instances to sparse format (Section 2.4)

Normalize Treat numeric attributes as a vector and normalize it to a given

length

Randomize Randomize the order of instances in a dataset

RemoveFolds Output a specified cross-validation fold for the dataset

RemoveMisclassified Remove instances incorrectly classified according to a specified

classifier—useful for removing outliers

RemovePercentage Remove a given percentage of a dataset

RemoveRange Remove a given range of instances from a dataset

RemoveWithValues Filter out instances with certain attribute values

Resample Produce a random subsample of a dataset, sampling with

replacement

SparseToNonSparse Convert all incoming sparse instances into nonsparse format