Project abandoned :(
Retrospection:
- The dataset collected is prone to heavy representational biases. Moreover, simplifying prediction of Likelihood of being admitted to a Graduate School based on Test Scores and Proxy such as GPA is a form of Measurement Bias. The data fails to reflect and/or consider multiple factors such as Net Finnancial Income, Deomgraphic,etc. Even then it would lend itself to Learning bias due to Representational/Sampling issues arising from data collected. EX: If factors such as age,gender, ethnicity are used then surely presence of a certain demographic will outweight others, this in-turn will lead to bias in our models by scoring similar candidates but with differrent backgrounds differently.
- Its been proven that factors such as GPA, Test Scores,etc are bad predictors of success both at finishing university as well as success outside university.
- A Complex model such as XGBoost/CatBoost will overfit and be biased. Simple estimators such as Linear Models or Decision Tree based models are better alternatives.
- From an ethical perspective the given dataset seems fit only for Statistical Analysis not for Predictive Analytics as it would lend to Deployment Biase.
Refer report for details such as architecture,methdology,etc.
A Streamlit❤️ web app that predicts the chance of admission into masters program based on various factors using a Flask API with Catboost model running as a background process on Windows .
Refer demo video.
python -m pip install -r requirements.txt
To run the API as a background process on Windows follow the instructions mentioned here
or one could open another console using tmux or in VSCode.
Run this file which serves as the entry-point to the Streamlit frontend.
streamlit run streamlit-app/streamlit_app.py
Following code boldy assumes the use of Windows. Since a Windows process is being spawned for serving the REST api. Also all file path need to be changed to match your dir structure.