A Reference Architecture for Educational Data Mining

In this paper we present a reference architecture for ETL stages of EDM and LA that works with different data formats and different extraction sites, ensuring privacy and making easier for new participants to enter into the process without demanding more computing power. Considering scenarios with a multitude of virtual environments hosting educational activities, accessible through a common infrastructure, we devised a reference model where data generated from interaction between users and among users and the environment itself, are selected, organized and stored in local "baskets". Local baskets are then collected and grouped in a global basket. Organization resources like item modeling are used in both levels of basket construction. Using this reference upon a client-server architectural style, a reference architecture was developed and has been used to carry out a project for an official foundation linked to Brazilian Ministry of Education, involving educational data mining and sharing of 100+ higher education institutions and their respective virtual environments. In this architecture, a client-collector inside each virtual environment collects information from database and event logs. This information along with definitions obtained from item models are used to build local baskets. A synchronization protocol keeps all item models synced with client-collectors and server-collectors generating global baskets. This approach has shown improvements on ETL like: parallel processing of items, economy on storage space and bandwidth, privacy assurance, better tenacity, and good scalability.