• Thanasekhar B
• Balaji C

Abstract: Decision trees trained in GPUs that followed batch processing overcame the insufficient memory to allocate the entire data by restricting the repeated computation of a split value in a node. This was achieved by implementation of the Node parallel processing method along with shared memory and synchronization among GPU cores. But this model suffered from a major load imbalance which leads to high communication cost. In order to minimize the imbalance in workload, a new model is proposed. The proposed model uses a histogram construction approach to compress the input data. This compression of data allows for faster transfer of data and less communication time between GPU cores. A Hybrid parallel architecture combining data and feature parallelism is proposed. Data parallelism is used in the compress function to compress the training data for computation. For split value calculation and entropy gain value calculation, Feature parallelism technique is used. This hybrid approach reduces the workload among worker nodes significantly thus leading to lower communication cost. The results of the proposed model with ijccn1 dataset indicate that it achieved a lower training time when compared with other parallel models. The evaluation of the proposed model against different datasets reflects the training performance is 10x speedup of the sequential model and 6x speedup of other models such as Sequential, OpenMP and OpenMPI models. The output of this phase can be used to determine the parameters which are significant in training the proposed model.

Decision trees, Batch Processing, Histogram construction, Data parallelism, openMP, MPI

"EFFICIENT PARALLEL PROCESSING ON DECISION TREES USING GPU", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.9, Issue 4, page no.c351-c360, April-2022, Available :http://www.jetir.org/papers/JETIR2204249.pdf

"EFFICIENT PARALLEL PROCESSING ON DECISION TREES USING GPU", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.9, Issue 4, page no. ppc351-c360, April-2022, Available at : http://www.jetir.org/papers/JETIR2204249.pdf