Resumen
Data mining is increasingly recognized as a key to analyzing, digesting and understanding the flood of digital data collected by business, government and scientific applications. Achieving this goal requires the scaling of mining algorithms to very large databases. Many classic mining algorithms require multiple database scans and random access to database records. Research as of August 1, 2002 focuses on overcoming limitations imposed when it is costly or impossible to scan large databases multiple times or access records at random. Predictive modeling is often a high-level goal of data mining in practice. There are two classes of algorithms namely decision tree methods and support vector machines. Decision trees are especially attractive in data mining environments since human analysts readily comprehend the resulting models. Their construction does not require an analyst to provide input parameters. Prior knowledge about the data is also not needed. Decision tree construction algorithms consist of two stages namely tree building and pruning. Support vector machines are other powerful and popular approaches to predictive modeling with success in a number of applications, including handwritten digit recognition, charmed quark detection, face detection and text categorization.