Dataset Characteristics (Metafeatures)

Summary This chapter discusses dataset characteristics that play a crucial role in many metalearning systems. Typically, they help to restrict the search given configuration space. The basic characteristic of target variable, for instance, determines choice right approach. If it is numeric, suggests suitable regression algorithm should be used, while if categorical, classification used instead. provides an overview different types characteristics, which are sometimes also referred as metafeatures. These types, and include so-called simple, statistical, information-theoretic, model-based, complexitybased, performance-based last group has advantage can easily defined any domain. include, sampling landmarkers representing performance particular algorithms on samples data, relative capturing differences or ratios values providing estimates gains . final part this specific machine learning tasks, including classification, regression, time series, clustering.