Constructing X-of-n Attributes With A Genetic Algorithm

摘要: The predictive accuracy obtained by a classification algorithm is strongly dependent on the quality of attributes data being mined. When are little relevant for predicting class record, will tend to be low. To combat this problem, natural approach consists constructing new out original attributes. Many attribute construction algorithms work simply conjunctions and/or disjunctions attribute-value pairs. This kind representation has limited expressiveness power represent interactions. A more expressive X-of-N [Zheng 1995]. An condition set N value an given example (record) number pairs that match with condition. For instance, consider following condition: X-of-{"Sex = male", "Age < 21", "Salary high"}. Suppose pairs: {"Sex 51", 2 3 condition, so 2.