A Hybrid Machine Learning Model for Grade Prediction in Online Engineering Education

Abstract

Imposed and exclusively online learning, caused by COVID-19, revealed research challenges, e.g. curricula reformation and data collection. With this pool of data, this research explores grade prediction in an engineering module. A hybrid model was constructed, based on 35 variables, filtered out of statistical analysis and shown to be strongly correlated to students' academic performance. The hybrid model initially involves a Generalized Linear Model. Its errors are used as an extra dependent variable, incorporated to an artificial neural network. The architecture of the neural network can be described by the sizes of the: input layer (36), hidden layer (1), output layer (1). Since new factors are revealed to affect students' academic achievements, the model was trained in the 70% of participants to forecast the grade of the remaining 30%. The model has therefore been divided into three subsets, with a training set of 70% of the sample and one hidden layer predicting the test set (15%) and the validation set (15%). Finally, the model has yielded an R2 of one. This suggests that the modeling framework effectively links the predictors with the grade (dependent variable) with absolute fitting success.