No abstract

Decision Tables is a well known classification algorithm which is both efficient and accurate. This paper presents the Parallel Scheme of Decision Tables (ParalTabs) which is an implementation of decision tables using the parallel model of Single Program and Multiple Data Streams (SPMD). This model communicates through shared memory, ie, the threads communicate with each other by reading and writing in the same physical address space. The algorithm uses a parallel scheme that follows the strategy of divide and conquer (D & C). Data is given to different threads to work on and the results collected to obtain the final decision table. We found, by a series of tests, the granularity most appropriate to divide data and obtain a reduction in execution times. A sequential version of Decision Tables was used to perform tests on the data and also other classification tools were used in order to have a thorough comparison with the parallel classifier proposed. We found ParalTabs a useful algorithm to perform classification on large databases, obtaining improvements in execution times and performance measures.

No abstract

No abstract

No abstract