This topic describes the binning methods supported by Oracle Data Miner for numerical and categorical attributes. For information about interval notation, see Interval Notation.
Oracle Data Miner supports these methods for binning numerical attributes:
min
is the minimum value of an attribute and max
is the maximum value and that you request N
quantile bins. The range [min, max] is divided into M
intervals, where M
<=N
, in such a way that each interval contains the same number of attributes. For detailed information about how many bins are generated, see Actual Number of Bins in Quantile Binning.min
is the minimum value of an attribute and max
is the maximum value and that you request N
equal-width bins. The range [min, max] is divided into N
equal intervals of length d
, where d
equals ((max-min)/N)
). Bin 1 is [min, min+d)
, bin 2 is [min+d, min+2d)
, and bin N
is [min+(N-1)*d,max]
. In general, different bins will contain different numbers of attributes.Quantile binning creates bins with approximately equal numbers of cases in each bin, irrespective of the width of the numerical range. Equal Width binning creates bins of identical width, irrespective of the number of cases in each bin; note that this strategy could generate empty bins.
Oracle Data Miner supports this method for binning categorical attributes:
N
bins. The wizard determines the most frequent N
values. The attributes are divided into N
+1 bins, where bin 1 contains attributes with frequent value number 1, ... bin N
contains attributes with frequent value number N
, and bin N
+1, the "other" bin, contains those attributes that take on none of the N
frequent values.There is also a binning strategy of None for categorical attributes, that is, do not bin the attribute.
Copyright © 2006, 2008, Oracle. All rights reserved.