Investigating to what extent learning hours affect students’ GPA through a comparison of linear regression and neural-network machine learning models

This paper investigates that to what extent students’ study time can be used to predict their Grade Point Average (GPA) through the application of machine learning algorithms. As the relevance of predicting academic performance continues to rise in the field of education, the use of data analysis offers a practical means of exploring the relationship between learning behavior and academic performance. A publicly accessible dataset with study time and corresponding GPA data was employed for analysis. Two models were employed and compared for predictive purposes: linear regression and a neural network model. Linear regression was employed as a control model to establish the linear relationship between study time and GPA, while the neural network model was employed to explore the possibility of non-linear relationships within the data. Standardization of the data was conducted before model training, and the performance of the models was assessed using Mean Absolute Error (MAE) and Mean Relative Error (MRE), as well as graphical assessment using scatter plots. The results show that although linear regression establishes a straightforward and easy-to-interpret relationship, the neural network model performs better in terms of predictive accuracy by exploring more complex relationships.