Ensemble pruning aims to achieve a good classification result by using smaller classifiers by finding the optimal subset. Classifier diversity and accuracy are widely recognized as two key factors for the success of an ensemble. There is a trade-off between classifier diversity and accuracy, which allows the ensemble to achieve the best performance. Existing ensemble pruning methods typically consider diversity and accuracy separately, and despite extensive work in this field, dynamic pruning of Random Forests is underexplored. This paper presents a dynamic pruning approach for decision forests that considers both diversity and forest effectiveness simultaneously, using thresholds to decide whether an ensemble of trees is accepted into the forest.