The use of machine learning “black boxes” explanation systems to improve the quality of school education

Abstract

The paper describes development of a multi-criteria decision support system (MCDSS) to improve the quality of school education. It is proposed to apply interpretable machine learning models for making decisions on improving the quality of education in secondary schools. Existing DSS are based on the expert judgement, which can be subjective. In addition, the large amount of data and features makes manual analysis difficult. Our approach is referred to as MCDSS with “black boxes” explainer, it consists of three stages. First, we develop the target indicators that measure the quality of education. A set of four features of quality of education (Q-Edu) has been developed. Secondly, we build regression models that link the data of the national educational database (NEDB) with target indicators. Thirdly, we use machine learning model interpreters to develop recommendations. The disadvantage associated with the difficulties of interpreting the results of models is overcome by SHAP (SHapley Additive exPlanations), which is used as a basis for developing recommendations for what features of educational institution could be altered in order to improve quality indicators. Using the described process, we, in particular, revealed the positive impact of the location of the school, ratio of experienced teachers, sports, technical and art studios on Q-Edu indicators. The ratio of experienced teachers and, at the same time, young teachers younger than 25 year positively affects the number of significant student achievements. The proposed universal approach reduces the subjectivity and laboriousness of parameter significance determination in MCDSS.