Experience with Model Sharing in Data Mining Environments

The most common case of interoperability in data mining environment deals with model sharing. A model in the scope of data mining represents the pertinent information which is extracted from the raw data using a suitable algorithm. The purpose of this paper deals with a strategy of exchanging models between data mining environments based on the definition of a canonical representation. The choice of PMML to serve this need could ease the diffusion of the suggested solution. While the commercial and the budding tools incorporated the PMML schema, the research environments in the domain focus on the performance of the algorithms. That is why we designed generic adapters which are responsible to transform the native structure of the model extracted within these environments towards the canonical representation which is considered in this study. The goal is being to avoid rewriting the codes of the existing algorithms.