Predictive Analytics for Early Identification of At-risk Students
- In the realm of education, the timely identification of students who need further support to succeed in their respective courses, plays a pivotal role in fostering aca- demic success and preventing potential setbacks. This thesis thus aims to contribute to this critical area by focusing on the development of predictive models for the early detection of at-risk students in their academic journey. The primary dataset used for this thesis is provided by kaggle, encompassing diverse student informa- tion, including demographic, socio-economic factors, and academic performance categorized into three different classes, presenting an imbalanced nature that poses a significant challenge.
Thus the primary objectives of this thesis are to address the problem of imbal- anced data, explore and assess the performance of multiple classification methods such as, logistic regression, decision tress, random forests and support vector ma- chines (SVM), neural networks, and create a comprehensive end-to-end processing pipeline which includes the systematic steps of balancing the data, model training and evaluation. Additionally the developed pipeline is tested on two additional datasets to assess its generalizability and robustness. This research aims to provide a comprehensive understanding of addressing the challenges of imbalanced data and how different classification methods and regression can be optimally applied to early detection of at-risk students. The findings are expected to aid educational institutions in supporting their students and enhancing academic success through timely interventions.
Key findings demonstrates the robustness of SVM SMOTE balancing technique acro- ss the datasets used in this study, where it consistently achieved best results when combined with various models, particularly highlighting the success of the combi- nation of Random Forest model with SVM SMOTE, and Decision tree model with SVM SMOTE in achieving notable accuracy rates. This emphasizes the adaptability of the balancing techniques employed, providing a strong foundation for predictive intervention educational settings.
