This repository contains a streamlit application with enabled function calling , that predicts employees attrition using a xgboost machine learning model.
Below is the Demo showing , the app in action
The main notebook covers all the stages of cleaning, preprocessing and training the provided dataset.
The algorithm used here is XGBoost and achieved an F1 score of 0.9003, after extensive hyperparameter optimization and feature selection
I used F1 score because we are dealing with an imbalanced dataset , and we need a balanced way to evaluate the performance of the model
$F1 = 2 * (precision * recall) / (precision + recall)$
For this part, I made use of langchain to make the whole process very cohesive.
Also for the LLM , I am making use of the newly released Llama 3.1-70b model , it is a very versatile model, and has been trending a lot because of its spectaular performance on function calling.
To set this up locally, start by cloning the application
git clone https://github.com/eskayML/employee-attrition-predictionOnce you're inside the peepalytics-assessment directory after cloning, you can run
pip install -r requirements.txt
Locally, You will need to add a file named secrets.toml to the .streamlit/ folder , this would be where the api keys are stored securely for streamlit to access
for the secret key, you would need a Groq api key, you can get one at https://console.groq.com/keys
you insert your groq keys into the created secrets.toml file
GROQ_API_KEY = "your groq api key"Then finally go back to the terminal and execute
streamlit run app.py