Learning Analytics (LA) is an emerging practice at a very early stage of adoption especially inEurope [10]. In particular, LA, which incorporates Predictive Analytics (PA), is dependent on large and rich datasets for accuracy. Most LA solutions use data available from Learning Management Systems (LMS), and Student Information Systems (SIS). As such, the data used is limited to the student’s activities and interactions with these systems. This data might in turn be inadequate forbasing analysis and predictions on. Many LA solutions are therefore looking beyond these systemsto include data from other sources as well. Data such as social media analysis, library use, studentbehaviour based on access card swipes and IP address as well as other multimodal data from avariety of sensors etc could be included.

There are several motivations for integrating data such as scaling up LA projects, enabling improvements of new educational technologies, and making more accurate and fine-grained analyses based on a wider set of data.Recent research points to the limited focus on technical details of integrating data and the very limited use of data integration specifications [8]. This study,therefore, aims to help close this gap in current research by identifying limitations and issues in data integration based on previous research efforts to inform and propose an architecture to enable further development and opportunities when scaling up LA projects.

The proposed architecture enables data integration of different data types from various educational technologies used in the learning environment. Drawing on principles of clean architecture[6] the decoupling of business rules is informing the architecture design. As such a component is also independent of other external “layers” such as user interface, frameworks and database.