MAP - Charting Student Math Misunderstandings

Project Overview

This project uses Natural Language Processing (NLP) and Machine Learning to identify student math misconceptions from open-ended responses. It is based on the Kaggle competition dataset.

The project implements a 3-stage modeling approach:

1. Binary Classification: Predict correct vs. incorrect answers.
2. 3-Class Classification: Categorize explanations (Correct, Misconception, Neither).
3. Multiclass Classification: Identify specific misconception types (35+ categories).

Structure

• Term_Project_geissinger_final.ipynb: Main analysis and modeling notebook.
• project_math/: Folder containing the dataset (train.csv, test.csv).
• requirements.txt: List of dependencies.

Setup

1. Clone the repository.
2. Install dependencies:

   pip install -r requirements.txt

3. Run the notebook: Open Term_Project_geissinger_final.ipynb and run all cells. The notebook is configured to look for data in the project_math/ directory by default.

Models

• Text Representation: TF-IDF and Sentence Transformers (embeddings).
• Classifiers: Random Forest and Logistic Regression.
• Handling Imbalance: BorderlineSMOTE.