A New Local Causal Learning Algorithm Applied in Learning Analytics

Purpose: The paper aims to present a new two stage local causal learning algorithm -- HEISA. In the first stage, the algorithm discoveries the subset of features that better explains a target variable. During the second stage, computes the causal effect, using partial correlation, of each feature of the selected subset. Using this new algorithm, the study aims to identify the actions that lead a student succeed or failure in a course. Design/methodology/approach: The paper presents a brief review of main concepts used in this study: Causal Learning and Causal effects. The paper also discusses the results of applying the algorithm in education data set. Data used in this study was extracted from the log of actions of a Learning Management System, Moodle. These actions represent the behavior of 229 engineering students that take Algorithm and Data Structure course offered in a blended model. Findings: The algorithm proposed in the paper identifies that features with weak relevance to a target may become relevant when computing the direct effect. Research limitations/implications: The algorithm needs to be improved to automatically discard attributes that are under a specific threshold of direct effect. Researchers are also encouraged to test the proposed propositions further. Practical implications: The algorithm presented in this paper can be used to identify the mostly relevant features given a classification task. Originality/value: This paper computes the direct effect of a selected subset of features in a target variable to evaluate if a variable in this subset is really a cause of the target or if it is a spurious correlation.