Communication-less Strategies for the Widening of Rule Induction

In the age of Big Data and with the ever increasing availability of parallel compute resources there has been strong focus on research in parallel algorithms for data mining aiming to improve the efficiency of existing algorithms. We take a different view, instead of the usual focus on speed-up of the algorithm, we focus on investing parallel compute resources to improve the accuracy of models obtained by existing heuristics, without increasing the overall running time. We look for strategies to invest parallel compute resources in a smart way in order to improve the search space exploration, without the necessity of communication between the parallel workers. We demonstrate their effectiveness on the rule induction algorithm CN2.