As for improving the confidence, the best way to do this is getting more data. This will, in general, produce terminal nodes based on more examples, thereby tightening the aforementioned estimates. Additionally, you can improve the confidence level of your model by generating ensembles, engaging in some feature engineering, clustering your sources to find similarities among them before building separate classification or regression models per cluster, or removing anomalous instances in your dataset.
This blog post provides an extensive analysis of your predictive model with some good tips to improve the results.