• knowledge discovery
• data mining
• data streams
• classification

Mining data streams has recently become an important and challenging task for a wide range of applications, including sensor networks and web applications. The massive quantity of streaming data coupled with concept drifting are two crucial issues in mining data streams. This thesis proposes a new framework for data streams classification, introducing two distinct structures to face the problem of data management and mining. On the one hand, our approach provides a synthetic structure which maximizes data availability, guaranteeing a single data access. On the other, given the synthetic structure, a selective ensemble of classifiers is managed through time to provide a good prediction accuracy. Both components are designed to maximize data usage and accuracy even in the presence of concept drifting, providing a good trade-off between data access management and quality of the model.