SELECTION OF CORE PREDICTIVE FEATURES OF STUDENTS’ ACADEMIC GRADES USING BRIDGE ESTIMATION METHOD

Abstract

It is frequently a key concern of tertiary institution administration and academics in particular to understand the predetermining factors that are most related to student’s academic performance. This study tends to examine the efficiency of some sparse estimators (i.e., Group-lasso, Sparse Group-LASSO and Group-Bridge) in predicting the student’s final grades by selecting the core predictive factors. The grades of the students (Failed/Passed) are the target variable of interest in the data, and the predictors include seven (7) variables: gender, number of absence days, relations, programs, parent's educational background, Course units, and Grade points. The original set of data was split (30:70) into a training set of 2800 observations and a test set of 1200 observations. The three (3) estimators were assessed for classification and prediction of their probability of a given target variable, using the Sensitivity (SEN), Specificity (SPEC), Misclassification Error Rate (MER), Positive Predictive value (PPV), Negative Predictive value (NPV), and Precision (PREC). It was discoveredthat the Group-Bridge estimator selected Grade points, Gender, and Parental Education Background as the core relevant factors to the students’ academic achievement, while the baseline estimator (Group-LASSO) and Sparse Group-LASSO selected Subject units and Grade points. Also, the study found that Sparse Group-LASSO is much more efficient the other two estimators in selecting and predicting the core relevant predictors.