undersample_tomek

A Tomek link is a minority instance and majority instance that are each other's nearest neighbor. This function removes sufficient Tomek links that are an instance of cls to yield m instances of cls. If desired, samples are randomly discarded to yield m rows if insufficient Tomek links are in the data.

Imbalanced training datasets impede many popular classifiers. To balance training data, a combination of oversampling minority classes and undersampling majority classes is useful. This package implements the SCUT (SMOTE and Cluster-based Undersampling Technique) algorithm as described in Agrawal et. al. (2015) <doi:10.5220/0005595502260234>. Their paper uses model-based clustering and synthetic oversampling to balance multiclass training datasets, although other resampling methods are provided in this package.

Keenan Ganz

scutr

Balancing Multiclass Datasets for Classification Tasks

undersample_tomek function

<dl><dt>data</dt>
<dd>Dataset to be undersampled.</dd>
<dt>cls</dt>
<dd>Majority class to be undersampled.</dd>
<dt>cls_col</dt>
<dd>Column in data containing class memberships.</dd>
<dt>m</dt>
<dd>Desired number of samples in undersampled dataset.</dd>
<dt>tomek</dt>
<dd>Definition used to determine if a point is considered a minority in the Tomek link definition.<ul>
<li><code>minor</code>: Minor classes are all those with fewer than <code>m</code> instances.</li>
<li><code>diff</code>: Minor classes are all those that aren't <code>cls</code>.</li>
</ul></dd>
<dt>force_m</dt>
<dd>If <code>TRUE</code>, uses random undersampling to discard samples if insufficient Tomek links are present to yield <code>m</code> rows of data.</dd>
<dt>...</dt>
<dd>Additional arguments passed to <code><a href="/link/dist?package=scutr&version=0.2.0" data-mini-rdoc="scutr::dist">dist</a>()</code>.</dd></dl>

Arguments

Undersample a dataset by removing Tomek links. — undersample_tomek

<dl>

<dt>data</dt>
<dd>Dataset to be undersampled.</dd>


<dt>cls</dt>
<dd>Majority class to be undersampled.</dd>


<dt>cls_col</dt>
<dd>Column in data containing class memberships.</dd>


<dt>m</dt>
<dd>Desired number of samples in undersampled dataset.</dd>


<dt>tomek</dt>
<dd>Definition used to determine if a point is considered a minority in the Tomek link definition.<ul>
<li><code>minor</code>: Minor classes are all those with fewer than <code>m</code> instances.</li>
<li><code>diff</code>: Minor classes are all those that aren't <code>cls</code>.</li>
</ul></dd>


<dt>force_m</dt>
<dd>If <code>TRUE</code>, uses random undersampling to discard samples if insufficient Tomek links are present to yield <code>m</code> rows of data.</dd>


<dt>...</dt>
<dd>Additional arguments passed to <code><a href='https://rdrr.io/r/stats/dist.html'>dist</a>()</code>.</dd>

</dl>

undersample_tomek: Undersample a dataset by removing Tomek links.

Description

Usage

Value

Arguments

Examples