Strategies to improve variable selection performance

The increasingly large datasets available for data mining and machine learning task are placing a premium on algorithm performance. The datasets are increasing in both rows (i.e., records) and columns (i.e., variables). One critical item that impacts the performance of these algorithms is the approach taken for storing and processing the data elements. This paper examines the performance tradeoffs between row major order and column major order in the context of heuristic variable selection for the case when variable selection is frequently used in some applications. The column major order approach is empirically shown to provide better runtime performance than the row major order.