JacobWeinbren / Israeli-Regression

Israeli Demographic Election Model

This codebase is structured to handle data processing, model training, and prediction for machine learning on the Israeli Election Study 2022. It includes scripts for data preparation, model training, evaluation, and a web application for making predictions.

Key Components

Data Preparation and Analysis

• Data Loading and Filtering: Data is loaded from SPSS .sav files using pyreadstat and filtered based on specific criteria.

python scripts/main.py

• Class Distribution Analysis: Analyses the class distribution for categorical variables.

python scripts/distribution.py

Model Training and Evaluation

• Pipeline Creation: A pipeline is created for preprocessing and model training using XGBoost and RandomForest within a VotingClassifier.

• Hyperparameter Optimization: Optuna is used for hyperparameter optimisation of the XGBoost model.

• Model Evaluation: Models are evaluated using ROC AUC scores for both One-vs-Rest (OVR) and One-vs-One (OVO) strategies.

• Prediction Script: A standalone script for making predictions from the command line.

python scripts/prediction.py

Frontend

• Astro-based Frontend: The frontend is built using Astro, using components for form inputs and displaying prediction results.

cd site
npm run dev

Dependencies

The project relies on several key Python libraries, including Flask, joblib, imblearn, xgboost, pandas, Flask-Cors, dask-ml, and Gunicorn, as specified in requirements.txt.

License

The project is licensed under the MIT License, allowing for wide use and modification.