Research on incremental learning on cloud computing platform

Research on incremental learning on cloud computing platform

Nanjing University of Posts and Telecommunications Li

Combining with the characteristics of traditional incremental learning algorithm, after deeply analyzing the MapReduce framework mechanism, this paper designs two incremental classification algorithms based on cloud computing platform in the process of integrating modular integrated learning into incremental learning. The map phase of the two algorithms is the training of the corresponding base classifier, and the different map tasks can be executed in a high degree of parallelism. Reduce stage is based on whether there is a concept drift phenomenon in the learning environment, using the classifier combination method and classifier selection method to integrate the map phase base classifier, which realizes the fast and efficient incremental data mining task without concept drift and concept drift environment respectively. The simulation experiments on KDD2010, hyperplane and other datasets show the correctness and feasibility of the proposed algorithm.

Keywords: incremental classification concept Drift cloud computing Hadoop

[Download Address]:http://bbs.chinacloud.cn/showtopic-13309.aspx