Importance and interactions of attributes in classification

This masters degree provides the in-depth look of the relations between attribute importance and interactions between machine learning attributes. The importance of the attribute was calculated with information gain and ReliefF while interactions between attributes were calculated with the use of information gain. Using Java and Weka we implemented an environment in which we empirically analyzed 63 known examples. We concluded that attribute importance and attribute clarity were strongly related. It was also concluded that if we wish to know the majority of interactions between attributes, great attribute knowledge is required. This is against the desired effect which is that for knowing the majority of interactions a small number of the most important attributes is required. Attributes linked with stronger interactions also have a bigger number of positive interactions