Prediction of Student Performance in Engineering Drawing Using Machine Learning Methods and Synthetic Minority Oversampling Technique (SMOTE)

Abstract

Engineering drawing courses are very crucial to engineering students forming the bases for more advanced design engineering courses. Seeing the high failure rate of engineering students in Nigerian higher institutes for various reasons. This research paper predicted the performance of students in engineering drawing courses at introductory level. Data was collected using paper-based questionnaire from engineering students in different departments. Logistics regression classifier and decision tree machine-learning algorithms were employed. SMOTE was introduced into the training phase in a bid to improve the prediction accuracy of the model. The machine-learning model was built on Kaggle with tools from Python Kernel. After running the models on a testing data, all the models were capable of classifying a successful outcome with accuracy between 67% - 78%. Logistics regression had the highest chance of prediction. The introduction of SMOTE clearly improved the prediction rate.