Scalable Feature Selection for Large Sized Databases

Feature selection determines relevant features in the data. It is often applied in pattern classiica-tion. A special constraint for feature selection nowadays is that the size of a database is normally very large. An eeective method is needed to accommodate the practical demands. A scalable probabilistic algorithm is presented here as an alternative to the exhaustive and heuristics approaches. The scalable probabilistic algorithm is designed and implemented to meet the needs arising from real-world data mining applications. Through experiments, we show that (1) the probabilistic algorithm is eeective in obtaining optimal/suboptimal subsets of features; (2) its scalable version expedites feature selection further and can scale up without sacriicing the quality of selected features.