Process mining refers to the extraction of process models from event logs. As real-life processes tend to be less structured and more flexible, clustering techniques are used to divide traces into clusters, such that similar types of behavior are grouped in the cluster. Educational process mining is an emerging field in the educational data mining (EDM) discipline, concerned with developing methods to better understand students' learning habits and the factors influencing their performance. However, the obtained models, usually, cannot fit well to the general students' behaviour and can be too large and complex for use or analysis by an instructor. These models are called spaghetti models. In the present work, we propose to use a two steps-based approach of clustering to improve educational process mining. The first step consist of creating clusters based employability indicators and the second step consist on clustering the obtained clusters using the AXOR algorithm which is based on traces profiles in order to refine the obtained results from the first step. We have experimented this approach using the tool ProM Framework and we have found that this approach optimizes at the same time, both the performance/suitability and comprehensibility/size of the obtained model.