4.2.4. Model Predictions

After the models were trained on the available data, the models can compute or predict corresponding target values for the feature variables that were previously unknown to the model. Unknown feature values are equally distributed data points from a specified interval as well as the features of a test data set. For the former, the minima and maxima of the feature values of the training data set were extracted. Afterwards, equally distributed data points were generated for each feature in this min-max interval.

For predicting the targets, the scikit-learn library provides the method predict(x), where the feature variables are passed as a vector x to the function. Calling the method reg\_model.predict(x) then returns the corresponding predicted target values. The predictions for the test data were further evaluated in terms of the root mean squared error, the mean absolute error and the coefficient of prognosis (CoP) [53] and showed good quality, especially for the GPR model (see Table 3).


**Table 3.** Prediction quality of the models based on the initial dataset.

From Table 3, it follows that the GPR model is the most suitable model for further evaluation in our test case since it shows the highest coefficient of prognosis. Therefore, we selected the GPR model for the demonstration and visualization of our use case.
