MapReduce-based Data Cube Computation for Adverse Drug Reaction Analysis

• Main Authors: WANG, MIN-HSIEN, 王敏賢
• Other Authors: LIN, WEN-YANG
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/81730382862532171971

碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 104 === Recently, the detection of suspected ADR signals from the SRSs (spontaneous reporting systems) has been recognized as a useful paradigm for earlier discovery of unknown ADRs. The process however is quite expensive. The analysts have repeatedly to try different measures, parameter settings, and demographic factors such as Sex, Age, Race, etc. In view of this, we have used published FAERS data released by FDA to develop a web-based interactive system, named iADRs, to provide an OLAP-like analysis interface such that the analyst can interactively change signal measures and demographic factors. The kernel repository to iADRs is a new type of data cube named contingency cube which requires lots of computations and I/O overhead to construct. In this thesis, we propose a two-phase MapReduce-based framework to compute ADR contingency cube, with Phase 1 responsible for generating an intermediate cube used in Phase 2 to compute the required contingency cube. Experimental results conducted over the FAERS data show the great efficiency of our proposed method, which can complete the computation within half an hour on a Hadoop cluster composed of only three nodes.