Predicting Student Achievement: Exploring Non-Cognitive Feature Interactions Using Machine Learning Models

This research investigates how non-cognitive skills can predict student achievement, as measured by GPA. Non-cognitive traits like self-control, goal attainment, interpersonal connections, and leadership skills develop in students at various stages and are influenced, whether positively or negatively, by their environment and social circle. Because non-cognitive features alone are complex and intertwined, feature engineering is needed to create new features that combine these non-cognitive traits with each other or with cognitive features, in order to better predict student success by Analyzing their impact on academic performance at the end of the year. Various machine learning models including linear regression, gradient boosted regression model, random forest and XGBoost were employed and developed to assess the impact of these features. An important part of our approach includes feature engineering, which entails developing new features that incorporate the effects of both noncognitive features and, at times, cognitive and noncognitive features on student performance. Our findings show that the linear regression model performs the best while The Gradient Boosting and XGBoost models also have strong scores of 0.796 and 0.826, indicating a good fit to the data. These findings underline the significance of thoroughly studying non-cognitive factors on a large scale to establish connections between non-cognitive traits and cognitive traits, enabling the prediction of students' academic performance and early intervention for struggling students to encourage increased effort.