get_predictions

Function which makes a prediction for multinomial/logistic regression based on the given cut-off value and probabilities.

Predicts categorical or continuous outcomes while concentrating on a number of key points. These are Cross-validation, Accuracy, Regression and Rule of Ten or "one in ten rule" (CARRoT), and, in addition to it R-squared statistics, prior knowledge on the dataset etc. It performs the cross-validation specified number of times by partitioning the input into training and test set and fitting linear/multinomial/binary regression models to the training set. All regression models satisfying chosen constraints are fitted and the ones with the best predictive power are given as an output. Best predictive power is understood as highest accuracy in case of binary/multinomial outcomes, smallest absolute and relative errors in case of continuous outcomes. For binary case there is also an option of finding a regression model which gives the highest AUROC (Area Under Receiver Operating Curve) value. The option of parallel toolbox is also available. Methods are described in Peduzzi et al. (1996) <doi:10.1016/S0895-4356(96)00236-3> , Rhemtulla et al. (2012) <doi:10.1037/a0029315>, Riley et al. (2018) <doi:10.1002/sim.7993>, Riley et al. (2019) <doi:10.1002/sim.7992>.

Alina Bazarova

CARRoT

Predicting Categorical and Continuous Outcomes Using One in Ten
Rule

Marko Raseta

get_predictions function

<dl><dt>p</dt>
<dd>probabilities of the outcomes for the test set given either by an array (logistic regression) or by a matrix (multinomial regression)</dd>
<dt>k</dt>
<dd>size of the test set</dd>
<dt>cutoff</dt>
<dd>cut-off value of the probability</dd>
<dt>cmode</dt>
<dd><code>'det'</code> or <code>''</code>; <code>'det'</code> always predicts the more likely outcome as determined by the odds ratio; <code>''</code> predicts certain outcome with probability corresponding to its odds ratio (more conservative). Option available for multinomial/logistic regression</dd>
<dt>mode</dt>
<dd><code>'binary'</code> (logistic regression), <code>'multin'</code> (multinomial regression)</dd></dl>

Arguments

Predictions for multinomial regression — get_predictions

<dl>

<dt>p</dt>
<dd>probabilities of the outcomes for the test set given either by an array (logistic regression) or by a matrix (multinomial regression)</dd>


<dt>k</dt>
<dd>size of the test set</dd>


<dt>cutoff</dt>
<dd>cut-off value of the probability</dd>


<dt>cmode</dt>
<dd><code>'det'</code> or <code>''</code>; <code>'det'</code> always predicts the more likely outcome as determined by the odds ratio; <code>''</code> predicts certain outcome with probability corresponding to its odds ratio (more conservative). Option available for multinomial/logistic regression</dd>


<dt>mode</dt>
<dd><code>'binary'</code> (logistic regression), <code>'multin'</code> (multinomial regression)</dd>

</dl>

get_predictions: Predictions for multinomial regression

Description

Usage

Value

Arguments

See Also

Examples