Predicting Academic Success: A Foundational Step Towards Personalized Learning

This project demonstrates the potential of applying machine learning (ML) to the educational sector, given appropriate data. By training 14 models on 4 distinct datasets, we provide this project as a proof of concept for personalized learning through the prediction of academic success.

Overview

Utilizing four comprehensive datasets from the UCI Machine Learning repository, this study explores the prediction of students' academic success. These datasets cover a wide range of factors, including academic, social, socio-economic, demographic, and enrollment data, offering a holistic approach to understanding student performance.

Datasets and Training

• Data Split: 80% training, 20% testing
• ML Algorithms Explored:
  • Logistic Regression
  • Naïve Bayes
  • Decision Tree
  • Extreme Gradient Boosting (XGBoost)
  • Light Gradient Boosting (LightGBM)
  • AdaBoost
  • Random Forest
  • Stochastic Gradient Descent Classifier (SGD)
  • Support Vector Classifier (SVC)
  • Neural Network
  • 2 Stacking Classifiers (with various base learners)
  • 2 Voting Classifiers (one soft, one hard)
• Hyperparameter Tuning: Utilized GridSearchCV for optimization

Key Findings

• Portuguese Higher Education Dataset: Best performance by Logistic Regression (92% across all metrics) for graduation/dropout prediction.
• Portuguese High School Datasets: Superior performance by SVC for pass/failure prediction in Mathematics (91%) and Portuguese language (95%).
• Turkish University Dataset: A stacking classifier with an ensemble of Random Forest, AdaBoost, and Logistic Regression as base learners, and Logistic Regression as a final learner, showed best performance (83%).

System Architecture

Our solution includes an interactive user interface built with Streamlit, enabling users to make predictions on student outcomes. Below is an architectural overview:

SQL Database Integration

An SQL database comprising 4 tables for the models ensures automatic loading of relevant data upon user selection.

User Interaction

Users can perform inferences on individual or multiple students, choosing either to select students from the database or manually input feature values (requires dataset familiarity).

Repository Structure

The repository is organized into several key folders:

• Predicting Graduation Dropout 4.4k: Contains data, notebooks, saved models (.joblib), and figures for dropout prediction.
• Student Pass Fail POR: Hosts model training for Mathematics and Portuguese subjects, including data, notebooks, models, and figures.
• Student Performance - Turkey: Includes materials for predicting academic performance in Turkey.
• GUI Folder: Contains the Streamlit-based GUI and database. To run locally:
  1. Install dependencies from requirements.txt.
  2. Execute streamlit run Home.py.

Accessing the Interface

The Streamlit application is accessible at https://academic-success.streamlit.app/.