Research and Development in Intelligent Systems XXXI

Advances in hardware and software technologies allow to capture streaming data. The area of Data Stream Mining (DSM) is concerned with the analysis of these vast amounts of data as it is generated in real-time. Data stream classification is one of the most important DSM techniques allowing to classify previously unseen data instances. Different to traditional classifiers for static data, data stream classifiers need to adapt to concept changes (concept drift) in the stream in real-time in order to reflect the most recent concept in the data as accurately as possible. A recent addition to the data stream classifier toolbox is eRules which induces and updates a set of expressive rules that can easily be interpreted by humans. However, like most rule-based data stream classifiers, eRules exhibits a poor computational performance when confronted with continuous attributes. In this work, we propose an approach to deal with continuous data effectively and accurately in rule-based classifiers by using the Gaussian distribution as heuristic for building rule terms on continuous attributes. We show on the example of eRules that incorporating our method for continuous attributes indeed speeds up the real-time rule induction process while maintaining a similar level of accuracy compared with the original eRules classifier. We termed this new version of eRules with our approach G-eRules.

The current study presents a technique that aims at improving stability of feature subset selection by means of a combined instance and feature weighting process. Both types of weights are based on margin concepts and can therefore be naturally interlaced. We report experiments performed on both synthetic and real data (including microarray data) showing improvements in selection stability at a similar level of prediction performance.

Learning in complex and complicated domains is fundamental to performing suitable and timely actions within them. The ability of chess masters to learn and recall huge numbers of board configurations to produce near-optimal actions provides evidence that chunking mechanisms are likely to underpin human learning. Cognitive theories based on chunking argue in favour for the notion of bounded rationality since relatively small chunks of information are learnt in comparison to the total information present in the environment. CHREST, a computational architecture that implements chunking theory, has previously been used to investigate learning in deterministic environments such as chess, where future states are solely dependent upon the actions of agents. In this paper, the CHREST architecture is implemented in agents situated in “Tileworld”, a stochastic environment whose future state depends on both the actions of agents and factors intrinsic to the environment which agents have no control over. The effects of bounding agents’ visual input on learning and performance in various scenarios where the complexity of Tileworld is altered is analysed using computer simulations. Our results show that interactions between independent variables are complex and have important implications for agents situated in stochastic environments where a balance must be struck between learning and performance.