In this article we describe an approach for the prediction of the run-time of jobs in heterogeneous environments that applies a meta-prediction algorithm working in multiple phases. For an efficient utilization of hardware resources, it is necessary to support schedulers with detailed information about the jobs that are going to be dispatched. One technique is to provide accurate forecasts of application run-times. A large number of current approaches focus on limited sets of prediction techniques (often linear ones) whereas most times only one is deployed on a dataset, ignoring the different characteristics that are "denoted" inherently by the data and therefore are not obvious. For that reason we are proposing an adaptive system for run-time prediction that offers a large set of different and differently parameterized predictors respectively, whereas only the appropriate prediction techniques for specifically filtered clusters of jobs are executed.