Evolutionary Feature Extraction to Infer Behavioral Patterns in Ambient Intelligence

Machine learning methods have been applied to infer activities of users. However, the small number of training samples and their primitive representation often complicates the learning task. In order to correctly infer inhabitant’s behavior a long time of observation and data collection is needed. This article suggests the use of MFE3/GA\(^{D\!R}\), an evolutionary constructive induction method. Constructive induction has been used to improve learning accuracy through transforming the primitive representation of data into a new one where regularities are more apparent. The use of MFE3/GA\(^{D\!R}\) is expected to improve the representation of data and behavior learning process in an intelligent environment. The results of the research show that by applying MFE3/GA\(^{D\!R}\) a standard learner needs considerably less data to correctly infer user’s behavior.