DBC-Forest: Deep forest with binning confidence screening

As a deep learning model, confidence screening forest (gcForestcs) has achieved great success in various applications. Compared with the traditional approach, gcForestcs effectively reduces high time cost by passing some instances high-confidence region directly to final stage. However, there is group of low accuracy region, which are called mis-partitioned instances. To find these instances, this paper proposes binning (DBC-Forest) packs all into bins based on their confidences. In way, more accurate can be passed stage, and performance improved. Experimental results show that DBC-Forest achieves highly predictions for same hyperparameters faster than other similar models achieve accuracy.