Towards mega-modeling: a walk through data analysis experiences

Mega-modeling is a viable approach to data analysis by using a bottom-up, inductive method. Mega-models are large, autonomous computing systems whose interfaces are described through a data-centric approach and whose execution behavior can be inspected. Mega-modeling aims at defining a comprehensive theory and technology of model construction model search, model fitness evaluation, model composition, model reuse and model evolution. The building block of mega-modeling is the mega-module, a software component capable of processing 'big data' for analytical purposes. Every mega-module performs a well identified computation, which can be considered a unitary transformation from inputs to outputs. Every mega-module can internally use data and patterns that are considered as invariant in the context of the computation, whose extension can be either local or global. Every mega-module exhibits a format that consists of three distinct phases of information processing, data preparation, analysis, and evaluation.