Genetic algorithms

Abstract

There are two primary aims in data mining: to predict the unknown (the outcome of an event or the future value of a variable of interest); and to describe the patterns in the data that enabled the prediction, in a way that is intelligible to humans. Prediction in this case takes one of two forms: classification and regression. Classification involves assigning examples to one of a given set of categories, for instance diagnosing whether or not someone has a particular medical condition, or determining an unknown pattern from the known dataset. Regression involves assigning a real-valued number to the prediction,often in a time series. The computer science discipline concerned with developing algorithms for automated data mining is known as machine learning. The development of techniques and tools with the ability to intelligently and automatically assist humans in analysing the mountains of data for nuggets of useful information (Fayyad et al., 1996a). Machine learning is increasingly becoming ubiquitous, from the categorization of news, stories by aggregators like Google News and the suggestion of songs by online services such as Pandora, to the analysis of shopping habits by supermarkets (the reason for that loyalty card in your wallet). Machine learning tools monitor all credit card transactions looking for patterns of fraud and are used to manage financial portfolios of the few companies who publicly describe their investment systems called LBS Capital Management, who have used a combination of expert systems, neural networks and genetic algorithms since 1993 (Fayyad et al., 1996b). In science, the uses of machine learning, range from image analysis in astronomy to medical diagnosis (e.g. the identification of discriminatory genes in gene expression profiles for the diagnosis of cancer