Ontology Design for Scientific Theories That Make Probabilistic Predictions

A multidimensional ontology design paradigm based on Aristotelian definitions provides knowledge structure necessary for testing scientific theories that make probabilistic predictions. These predictions can be used to evaluate the theories on available data and can be used for new cases. This semantic-science framework can also be motivated by starting with machine learning. It is assumed that the theories make predictions about individuals and relations and are part of statistical relational learning. The data and the learned theories are assumed to be persistent and the theories are built using prior knowledge and multiple heterogeneous data sources that can be compared with other theories. It is also assumed that probability is the appropriate form of prediction for scientific theories. Probabilistic predictions minimize the prediction error for most error measures and are required to make decisions.