Hardware Acceleration of Nonincremental Algorithms for the Induction of Decision Trees and Decision Tree Ensembles
Bogdan Z. Vukobratović · National Repository of Dissertations in Serbia · 2017
The thesis proposes novel full decision tree and decision tree ensemble induction algorithms EFTI and EEFTI, and various possibilities for their implementations are explored. The experiments show that the proposed EFTI algorithm is able to infer much smaller DTs on average, without the significant loss in accuracy, when compared to the top-down incremental DT inducers. On the other hand, when compared to other full tree induction algorithms, it was able to produce more accurate DTs, with similar sizes, in shorter times. Also, the hardware architectures for acceleration of these algorithms (EFTIP and EEFTIP) are proposed and it is shown in experiments that they can offer substantial speedups.