Ventura, Dan, "On Discretization as a Preprocessing Step for Supervised Learning Models", Masters Thesis, Brigham Young University, April, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Salzberg [9] and used in the PEBLS system [10] and modifies the weighting scheme used by the VDM. These distance metrics work well in many symbolic domains, but they do not handle continuous attributes directly. Instead, they rely upon discretization, which often degrades generalization accuracy [11]. This paper presents two extensions of the Value Difference Metric which allow for more appropriate use of continuous attributes. Section 2 provides more background on the original VDM and subsequent extensions to it. Section 3 introduces the Interpolated Value Difference Metric (IVDM) and ....

....problem, as explored by Domingos [12] and Wilson Martinez [13] is to use a heterogeneous distance function. A heterogeneous distance function can use a different distance metric for each attribute, based on what kind of attribute it is [14] Another approach to this problem is discretization [11][15] Some models that have used the VDM or extensions of it (notably PEBLS [9] 10] have discretized continuous attributes into a somewhat arbitrary number of discrete ranges, and then treated these values as symbolic (discrete unordered) values. This method has the advantage of generating a ....

[Article contains additional citation context not shown here]

Ventura, Dan, (1995). On Discretization as a Preprocessing Step for Supervised Learning Models, Master's Thesis, Brigham Young University, 1995.

....study and sections four and five analyze these results and draw conclusions. 2. Description of Discretization Methods and Supervised Learners Used The discretization methods used are ChiMerge [10] equal width intervals[21] equalfrequency intervals [21] maxi min into k means [8] Valley [19][20], and Slice [20] This is by no means an exhaustive list of possible discretization techniques (see also [6] 9] 11] and [18] but it is believed to be a representative sample. ChiMerge. ChiMerge is based on the statistical c 2 test. All examples are sorted and initially each is placed in ....

....four and five analyze these results and draw conclusions. 2. Description of Discretization Methods and Supervised Learners Used The discretization methods used are ChiMerge [10] equal width intervals[21] equalfrequency intervals [21] maxi min into k means [8] Valley [19] 20] and Slice [20]. This is by no means an exhaustive list of possible discretization techniques (see also [6] 9] 11] and [18] but it is believed to be a representative sample. ChiMerge. ChiMerge is based on the statistical c 2 test. All examples are sorted and initially each is placed in its own ....

Ventura, Dan, "On Discretization as a Preprocessing Step for Supervised Learning Models", Masters Thesis, Brigham Young University, April, 1995.

....being parsimonious in its hypothesis representation. Future research includes extending MSP to handle nominal data (some related work has been presented in [1] 7] 15] investigating other metrics for prototype creation, and improving a parallel implementation based upon a c ary tree presented in [17]. Parallel implementation of MSP using a c ary tree with a broadcast and gather scheme is presented in [17] This work was funded in part by a grant from Novell, Inc. 6. ....

.... data (some related work has been presented in [1] 7] 15] investigating other metrics for prototype creation, and improving a parallel implementation based upon a c ary tree presented in [17] Parallel implementation of MSP using a c ary tree with a broadcast and gather scheme is presented in [17]. This work was funded in part by a grant from Novell, Inc. 6. ....

Ventura, Dan, "On Discretization as a Preprocessing Step for Supervised Learning Models", Masters Thesis, Brigham Young University, April, 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC