Method for computer-implemented determination of a data-driven prediction model

An apparatus for computer-implemented determination of a data-driven prediction model, comprising:

• a processor comprising hardware, the processor configured to:
• a) provide digital input data having data sets associated with a plurality of input variables and a number of output variables, the input data being semantically annotated based on a digital semantic representation, the semantic representation comprising a plurality of trees, where each tree is assigned to a respective input variable of at least some of the plurality of input variables and where each tree comprises a plurality of nodes representing ontology elements of a knowledge base, the nodes comprising a root node in an uppermost hierarchical level and nodes in one or more hierarchical levels lower than the uppermost hierarchical level, where each ontology element of a respective node in a hierarchical level is a sub-category of the ontology element of the node in the adjacent higher hierarchical level connected by an edge to the respective node, the nodes in the lowest hierarchical level being leaf nodes associated with discrete values of the input variable to which the respective tree is assigned;
• b) automatically recode the discrete values of the input variable to which each tree is assigned by determining a plurality of modified trees for the respective tree, a modified tree being derived by cutting off one or more hierarchical levels from the respective tree, thus substituting the leaf nodes of the respective tree by nodes in a higher hierarchical level, wherein each modified tree of the plurality of modified trees corresponds to a different recoding of the discrete values of the respective input variable to which the respective tree is assigned, wherein the plurality of modified trees comprises for each tree all modified trees derivable therefrom by cutting off one or more hierarchical levels;
• c) determine a plurality of different data modifications of the input data, each data modification comprising one or more recoded discrete values of one or more input variables, wherein the plurality of data modifications comprises all possible combinations of recoded discrete values based on the plurality of modified trees;
• d) train a machine learning method for predicting the number of output variables based on the plurality of input variables, the training being performed based on the data modifications, resulting in several trained machine learning methods;
• and
• e) determine the prediction quality of the trained machine learning methods, where the trained machine learning method having the highest prediction quality forms the determined data-driven prediction mode.

A computer program product, comprising a computer readable hardware storage device having computer readable program code stored therein, said program code executable by a processor of a computer system to implement a method when the program code is executed on a computer, the method including

• a) providing digital input data having data sets associated with a plurality of input variables and a number of output variables, the input data being semantically annotated based on a digital semantic representation, the semantic representation comprising a plurality of trees, where each tree is assigned to a respective input variable of at least some of the plurality of input variables and where each tree comprises a plurality of nodes representing ontology elements of a knowledge base, the nodes comprising a root node in an uppermost hierarchical level and nodes in one or more hierarchical levels lower than the uppermost hierarchical level, where each ontology element of a respective node in a hierarchical level is a sub-category of the ontology element of the node in the adjacent higher hierarchical level connected by an edge to the respective node, the nodes in the lowest hierarchical level being leaf nodes associated with discrete values of the input variable to which the respective tree is assigned;
• b) automatically recoding the discrete values of the input variable to which each tree is assigned by determining a plurality of modified trees for the respective tree, a modified tree being derived by cutting off one or more hierarchical levels from the respective tree, thus substituting the leaf nodes of the respective tree by nodes in a higher hierarchical level, wherein each modified tree of the plurality of modified trees corresponds to a different recoding of the discrete values of the respective input variable to which the respective tree is assigned, wherein the plurality of modified trees comprises for each tree all modified trees derivable therefrom by cutting off one or more hierarchical levels;
• c) determining a plurality of different data modifications of the input data, each data modification comprising one or more recoded discrete values of one or more input variables, wherein the plurality of data modifications comprises all possible combinations of recoded discrete values based on the plurality of modified trees;
• d) training a machine learning method for predicting the number of output variables based on the plurality of input variables, the training being performed based on the data modifications, resulting in several trained machine learning methods;
• e) determining the prediction quality of the trained machine learning methods, where the trained machine learning method having the highest prediction quality forms the determined data-driven prediction model.