Published Paper Details:

Decision Tree classification is one of the most widely used machine learning algorithm in dozens of sectors due to its classification effectiveness and model interpretability. However, the voluminous training data and the inherent problem complexity are the two major factors that strongly influence the time and computational efficiencies of the decision tree classification models. As the complexity of the problem increases, the number of computations required to learn every significant pattern will also increase. These enormous computations upsurge the training time and necessitate extensive computing resources especially when the training dataset is huge. To analyze such voluminous and complex datasets at a rapid speed, the need to shift to parallel processing approaches is essential for lowering overall computational costs. In this scenario, this paper provides a comparative analysis of various parallel implementations of decision tree classification models in Apache Spark on three UCI big datasets namely SUSY, PokerHand, and RLCP. The performance exhibited by various parallel implementations is compared in terms of scalability, time complexity and classification performances.