Application of Machine Learning Models for Predicting Students' Performance in Mathematics: A K-Fold Approach

Mathematics plays a pivotal role in science, technology and engineering. It is one of the compulsory subjects to gain admission into a higher institution in Nigeria. Considering its importance, students' performance has been a significant concern for most educators, government, parents and guardians. This study aims to predict students' performance in mathematics using the K-fold approach to support effective decision-making and policy implementation on mathematics. The primary dataset was collected from 192 senior secondary schools, and one mathematics student was randomly selected from five Senior Secondary Schools in Bwari Area Council, Abuja, and used as a sample. Students' Mathematics Achievement Test (SMAT) was used as the instrument to collect data. The data was processed and modelled using the Python Programming language. Thus, five machine learning models were trained and evaluated to overcome overfitting using four-, five---, and six-fold K-Fold cross-validation techniques. Five folds performed best for all the models except for the Support Vector Regressor, which performed best with six folds. The evaluation of the models showed that the Light Gradient Boosting Model outperformed others with a Mean Absolute Error of 12.642 and a Root Mean Square Error of 15.894. The models were significantly influenced by the students' personal study time, friends' assistance, and the parents' educational background. It is therefore recommended that academic researchers, educational authorities in Nigeria, and relevant stakeholders use and improve on this model to predict the students' scores and assist them in achieving great performance in mathematics.