Logistic regression and random forest classification models were used to predict students at risk of receiving a failing grade (D or F) in an introductory mechanics class (N = 2061 students). The dataset consisted of both institutional and in-class variables gathered from the first two weeks of instruction, with the in-class variables (homework and clicker scores) found to be particularly important for the model’s classification accuracy. Additional methods such as model regularization and upsampling of the failing class through the synthetic minority oversampling technique (SMOTE) were found to slightly improve model performance. The best performing models achieved a balanced accuracy of around 80%. An apparent performance ceiling of around 87% was found by providing the model with the course test average, indicating that substantially increasing performance above 80% is unlikely with data available early in the semester.

Underline digital video library

Using Machine Learning to Predict Student Performance in Introductory Mechanics

Ethnobotanical survey & Covid 19 pandemic: A case study of Fumaria in Iran

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES