Anticipating Teachers’ Performance

Data mining faces multiple defies, with improving its results through the use of domain knowledge at the top. The advent of domain driven data mining brings new techniques to use that knowledge. Educational data mining is a favored domain to explore such tools, both due to the advances in the area and the existence of domain knowledge. In this work, we propose a new methodology for anticipating teachers’ performance based on the analysis of pedagogical surveys. Our approach combines classification and sequential pattern mining to identify a set of meta-patterns that can be used to enrich source data, and in this manner we better describe teachers and consequently improve classification models accuracy. The use of domain knowledge is used in two steps: in the discovery of frequent behaviors (through the use of constraints) and in the enrichment of original data. A case study on mining pedagogical surveys is presented, corroborating our argument, and showing a significant improvement in classification accuracy.