Scalable Parallel Algorithms ForPredictive Modelling

摘要: Data Mining applications have to deal with increasingly large data sets and complexity. Only algorithms which scale linearly size are feasible. We present parallel regression after a few initial scans of the compute predictive models for mining do not require further access data. In addition, we describe various ways dealing complexity (high dimensionality) Three methods presented three different ranges attribute numbers. They use ideas from finite element method based on penalised least squares fits using sparse grids additive intermediate very high dimensional Computational experiments confirm scalability both respect number processors.