Guaranteeing Correctness of Machine Learning Based Decision Making at Higher Educational Institutions

Abstract

State-of-the-art software technologies have enabled Higher Education Institutions to record and store large amounts of student data. Analyzing this large amount of data can facilitate the decision-making process. Decisions made in higher education institutions affect policies, strategies and actions that improve education quality. It can be argued that machine learning algorithms demonstrate a remarkable ability to recognize models and predict results based on data. However, machine learning outcomes were subject to bias and erroneous labelling. Consequently, to mitigate the impact of such errors in decision making, it is necessary to put in place mechanisms to correct decision rules. This article presents an approach to learn decision rules through supervised machine learning and proves the correctness of these rules with hierarchical Coloured Petri Nets. The use of formalism in the proposed methodology ensures that the decision rules are correct and comprehensible. Empirical results show that we improved correctness with 98.68% accuracy on decision rules. This research work contributes to the improvement of the decision-making process for the academic administration of Higher Education Institutes.

Keywords

• Decision making
• supervised machine learning
• decision trees
• formal verification
• formal methods
• formal modeling