modelSelection

Multiple model performance metrics are computed

A stacking solution for modeling imbalanced and severely skewed data. It automates the process of building homogeneous or heterogeneous stacked ensemble models by selecting "best" models according to different criteria. In doing so, it strategically searches for and selects diverse, high-performing base-learners to construct ensemble models optimized for skewed data. This package is particularly useful for addressing class imbalance in datasets, ensuring robust and effective model outcomes through advanced ensemble strategies which aim to stabilize the model, reduce its overfitting, and further improve its generalizability.

E. F. Haghish 

autoEnsemble

Automated Stacked Ensemble Classifier for Severe Class Imbalance

modelSelection function

<dl><dt>eval</dt>
<dd>an object of class <code>"ensemble.eval"</code> which is provided
by 'evaluate' function. this object is a data.frame, including
several performance metrics for the evaluated models.</dd>
<dt>family</dt>
<dd>model family. currently only <code>"binary"</code> classification models
are supported.</dd>
<dt>top_rank</dt>
<dd>numeric. what percentage of the top model should be selected?
the default value is top 1% models.</dd>
<dt>max</dt>
<dd>integer. specifies maximum number of models for each criteria to be extracted. the
default value is the <code>"top_rank"</code> percentage for each model selection
criteria.</dd>
<dt>model_selection_criteria</dt>
<dd>character, specifying the performance metrics that
should be taken into consideration for model selection. the default are
<code>"c('auc', 'aucpr', 'mcc', 'f2')"</code>. other possible criteria are
<code>"'f1point5', 'f3', 'f4', 'f5', 'kappa', 'mean_per_class_error', 'gini', 'accuracy'"</code>,
which are also provided by the <code>"evaluate"</code> function.</dd></dl>

Arguments

Author

Selects Diverse Top-Performing Models for Stacking an Ensemble Model — modelSelection

<dl>

<dt>eval</dt>
<dd>an object of class <code>"ensemble.eval"</code> which is provided
by 'evaluate' function. this object is a data.frame, including
several performance metrics for the evaluated models.</dd>


<dt>family</dt>
<dd>model family. currently only <code>"binary"</code> classification models
are supported.</dd>


<dt>top_rank</dt>
<dd>numeric. what percentage of the top model should be selected?
the default value is top 1% models.</dd>


<dt>max</dt>
<dd>integer. specifies maximum number of models for each criteria to be extracted. the
default value is the <code>"top_rank"</code> percentage for each model selection
criteria.</dd>


<dt>model_selection_criteria</dt>
<dd>character, specifying the performance metrics that
should be taken into consideration for model selection. the default are
<code>"c('auc', 'aucpr', 'mcc', 'f2')"</code>. other possible criteria are
<code>"'f1point5', 'f3', 'f4', 'f5', 'kappa', 'mean_per_class_error', 'gini', 'accuracy'"</code>,
which are also provided by the <code>"evaluate"</code> function.</dd>

</dl>

modelSelection: Selects Diverse Top-Performing Models for Stacking an Ensemble Model

Description

Usage

Value

Arguments

Author