| --- |
| language: |
| - en |
| - es |
| tags: |
| - education |
| - early-warning-system |
| - sklearn |
| - tabular-classification |
| - random-forest |
| --- |
| |
| # Student Dropout Predictor (Early Warning System) |
|
|
| ## Model Description |
| This model is a Gradient Boosting Classifier classifier designed to predict the likelihood of student dropout (academic churn) in a virtual learning environment. It serves as the core analytical engine for an Early Warning System (EWS) that triggers proactive generative AI interventions. |
|
|
| - **Model Type:** Random Forest Classifier |
| - **Task:** Tabular Classification (Binary: Dropout vs. Retained) |
| - **Language:** English/Spanish feature support |
| - **Interpretability:** Fully compatible with SHAP (SHapley Additive exPlanations) to extract individual risk factors. |
|
|
| ## Intended Use |
| This model is intended for educational institutions to identify at-risk students based on their telemetry and engagement data (clicks, VLE interactions, assignment submissions, and demographics). |
| It is designed to be used in conjunction with a generative AI agent (e.g., Claude 3.5 via AWS Bedrock) to draft personalized intervention emails based on the root causes of the predicted risk. |
|
|
| ## Training Data |
| The model was trained on an anonymized dataset comprising student demographics, historical academic performance, and daily interaction logs with the Virtual Learning Environment (VLE). |
|
|
| ## Features |
| The input features include, but are not limited to: |
| - `vle_clicks`: Total interactions with the digital platform. |
| - `attendance_days`: Active days in the system. |
| - `assessments_submitted`: Number of assignments handed in. |
| - Demographics: Age band, disability status, prior education level. |
|
|
| ## Metrics |
| *(Note: Update these metrics based on your final evaluation script)* |
| - **Accuracy:** 0.77 |
| - **F1-Score:** 0.78 |
| - **ROC-AUC:** 0.84 |
