The modeling module in CVA provides the possibility to automatically generate models from data. To find the best model CVA applies multiple (mostly nonlinear) modeling algorithms on the data because it is impossible to determine the best one beforehand.

CVA optimizes the learning parameter for all methods through so-called hyperparameter optimization. This approach guarantees the configuration and selection of the best modeling algorithms and the generation of the best model. The target criterion for the models is their ability to generalize, i.e. the prediction quality of the models is evaluated. Using a so-called cross-validation approach avoids overfitting, a common phenomenon in modeling processes.

A traffic light system tells you how successful the modeling is and which model performs best. So you can directly proceed and apply the models for prediction, sensitivity analyses or optimization.

Experts have the possibility to influence the learning parameters of the methods. However, experience shows that this is rather unnecessary because of the excellent quality of the automatic modeling.

Although CVA provides all state-of-the-art methods of nonlinear data analysis, the generation of the optimal model is very easy for the user—only computation time is needed. The result is presented in an intelligible way. Among the available modeling algorithms are:

• Linear models
• Support vector machine
• Fuzzy models
• Decision trees
• Random forests
• Neural networks (MLP, feed-forward)
• Gaussian process
• Partial least squares regression
• Principal component regression