Interactive machine learning: letting users build classifiers

According to standard procedure, building a classifier using machine learni ng is a fully automated process that follows the preparation of training da ta by a domain expert. In contrast, interactive machine learning engages us ers in actually generating the classifier themselves. This offers a natural way of integrating background knowledge into the modelling stage-as long a s interactive tools can be designed that support efficient and effective co mmunication. This paper shows that appropriate techniques can empower users to create models that compete with classifiers built by state-of-the-art l earning algorithms. It demonstrates that users-even users who are not domai n experts-can often construct good classifiers, without any help from a lea rning algorithm, using a simple two-dimensional visual interface. Experimen ts on real data demonstrate that, not surprisingly, success hinges on the d omain: if a few attributes can support good predictions, users generate acc urate classifiers, whereas domains with many high-order attribute interacti ons favour standard machine learning techniques. We also present an artific ial example where domain knowledge allows an "expert user" to create a much more accurate model than automatic learning algorithms. These results indi cate that our system has the potential to produce highly accurate classifie rs in the hands of a domain expert who has a strong interest in the domain and therefore some insights into how to partition the data. Moreover, small expert-defined models offer the additional advantage that they will genera lly be more intelligible than those generated by automatic techniques. (C) 2001 Academic Press.