TLDR: This research introduces a Feedback-Driven Decision Support System (DSS) that uses a LightGBM model with incremental retraining to dynamically predict student performance. Unlike static models, this system continuously updates its predictions based on new data, such as post-intervention outcomes, leading to improved accuracy and more effective academic interventions. The system also features a web interface and SHAP for explainability, demonstrating a 10.7% reduction in prediction error after retraining.
Predicting how students will perform academically is crucial for providing timely help and support. However, many existing machine learning models used in education are static. This means they are trained once on old data and cannot adjust when new information, like how a student performs after receiving extra help, becomes available. This limitation makes them less effective over time because they don’t learn from real-world progress.
To tackle this challenge, researchers have proposed a new approach: a Feedback-Driven Decision Support System (DSS). This system features a unique ‘closed-loop’ design, allowing the prediction model to continuously improve itself. It integrates a powerful machine learning model called LightGBM, which is a type of regressor, with a mechanism for incremental retraining. This means that when educators input updated student results, the system automatically triggers an update to the model. This adaptive capability significantly enhances prediction accuracy by learning directly from actual academic progress.
The platform is designed with a user-friendly web interface, built using Flask, which allows for real-time interaction. To ensure transparency and build trust, the system also incorporates SHAP (SHapley Additive exPlanation) for explainability. This feature helps educators understand why the model makes certain predictions, showing which factors, such as study hours or attendance, are most influential.
Experimental results from testing this system have been very promising. After retraining the model with new feedback data, there was a notable 10.7% reduction in the Root Mean Squared Error (RMSE), a key metric for prediction accuracy. The R-squared value, which indicates how well the model explains the variation in student scores, also increased from 0.715 to 0.773. Furthermore, the system consistently showed upward adjustments in predicted scores for students who received interventions, demonstrating its ability to recognize and incorporate the positive impact of support.
This innovative system transforms traditional static predictors into self-improving tools. By doing so, it pushes educational analytics towards a more human-centered, data-driven, and responsive form of artificial intelligence. The framework is designed to be easily integrated into existing Learning Management Systems (LMS) and institutional dashboards, making it a practical solution for modern educational environments.
The core idea behind this DSS is its continuous cycle: predictions are made, interventions are applied, new performance data is collected as feedback, and then the model is retrained with this updated information. This ensures that the system remains relevant and accurate as student performance evolves. For more technical details, you can refer to the full research paper: Designing a Feedback-Driven Decision Support System for Dynamic Student Intervention.
The dataset used for this study, ‘Student Performance in Exams,’ was sourced from Kaggle and includes 6,607 student records with 19 input characteristics covering academic, behavioral, demographic, and environmental factors. Key features included ‘Tutoring Sessions,’ ‘Attendance,’ ‘Previous Scores,’ ‘Hours Studied,’ and ‘Parental Involvement,’ with ‘Exam Score’ as the target variable.
Also Read:
- Enhancing Decision Tree Performance Through Dynamic Pruning with Multi-Armed Bandits
- Holistic Explainable AI: Making AI Decisions Transparent for Everyone
In essence, this research bridges the gap between simply predicting student outcomes and actively supporting dynamic, real-time interventions. It represents a significant step forward in creating intelligent systems that learn and grow alongside the students they serve, fostering a more adaptive and effective educational ecosystem.
