Machine Learning and Deep Learning-Based Students’ Grade Prediction

Predicting student performance in a curriculum or program offers the prospect of improving academic outcomes. By using effective performance prediction methods, instructional leaders can allocate adequate resources and instruction more accurately. This paper aims to identify machine learning algorithm features for predicting student grades as an early intervention. Predictive models spot at-risk students early, allowing educators to provide timely support. Educators can customize teaching methods, and these models assess program success, helping institutions refine or expand them through data-driven decisions. But the problem definition of student grade prediction is to develop predictive models or algorithms that can forecast or estimate the future academic performance or grades of students based on various input features and historical data, and to do so, we utilized a student dataset comprising personal information and grades, employing various regression algorithms, including decision tree, random forest, linear regression, k-nearest neighbor, XGBoost, and deep neural network. We chose these algorithms for their suitability and distinct strengths. We assessed their performance using determination coefficient, mean average error, mean squared error, and root mean squared error. The results showed that the deep neural network outperformed others with a determination coefficient of 99.97%, confirming its reliability in predicting student grades and assessing performance, and this will certainly help to develop predictive models that can accurately forecast or estimate students’ academic performance based on various input features and enable teaching staff to provide timely assistance in addressing these issues.