Data-mining methods are aimed to derive hypothese from the data (section 1.1). Therefore data-mining results are also only hypothese and need to be taken with care as there are several dangers associated with ``misinterpreting'' a hypothese.
The general problem of causality, the relationship of wholes and parts, and the limiting factor of a training-set are three of these dangers that will be discussed in the following sections.