Ensemble-based factor analysis of survey data on computer science courses

Abstract

Analysis of course survey data are foundationally important because it has a close relation with students' skills training and talent development. However, there are multiple categories of classification information that characterized different students from different perspectives, such as genders, classes, types or regions. Compared to the correlation between survey data with individual kind of classification, consensus between those with multiple categories is more informative. However, traditional methods can not handle this problem with multiple kinds of classification categories. In view of the problem above, in this work, we proposed to conduct a new factor analysis method in the consensus fashion. Specifically, we computed a consensus structure of different categories of classifications and the similarity matrices for each factor, respectively. Motivated by the famous clustering measure RAND, we proposed a generalized method called Rmm which is used to compute the consistency of two different similarity matrices, with the option to adopt weights. Followed the similar spirit of Rmm, we also computed pairwise consistency between different factors without category information. Experimental results show that: (i) Our Rmm method is capable to combine different categories of classification information in an effective manner;(ii) our Rmm method preserve promising linear property for the combination of classifications; and (iii) Our Rmm method can be used in the case of fuzzy similarity, which is more applicable in different scenarios. With the proposed method, this work attempted to figure out the veiled laws and covered from read data, with the hope to provide suggestions for course arrangement and personnel cultivating program in universities, and to shed light on personal development of university students who major on computer science.