A cluster-based method for constructing sparse principal components is proposed. The method initially forms clusters of variables, using a new clustering approach called the semi-partition, in two steps. First, the variables are ordered sequentially according to a criterion involving the...