ML ASSIGNMENT 2 - Patient Emergency Risk Classification

TEAM 7

• Saharsh Misra - 2022A7PS0074H
• Aarav Haran - 2022B3A70880H
• Vaishnu Kanna - 2022B3A71608H
• Aditya Pentyala - 2022B3A70522H

USAGE INSTRUCTIONS

0. Ensure you have Python>=3.8 on your system
1. Clone this repository onto your local system
2. Download the dataset from the provided google drive link as part of the assignment and place it in the root directory of your clone (SKIP IF REPO DOWNLOADED WITH THE DATA DIRECTORY INTACT)
3. Create a new virtual environment with the command python -m venv env
4. Activate the environment:
   • source env/bin/activate on Mac/Linux
   • env/Scripts/activate.ps1 on Windows Powershell
   • env/Scripts/activate.bat on Windows cmd
5. Once inside the environment, install the requirements using pip install -r requirements.txt
6. Run the notebook eda.ipynb to populate the data/ directory (you may have to create an empty directory at first) (SKIP IF REPO DOWNLOADED WITH THE DATA DIRECTORY INTACT)
7. Run the command streamlit run dashboard.py

EVALUATION RUBRIC

1. Data Preprocessing and Exploratory Data Analysis (EDA) (5 Marks)
   • Done in eda.ipynb and through plots in dashboard.py
2. Feature Engineering and Representation (5 Marks)
   • Done in eda.ipynb
3. Classification Models (5 Marks)
   • Done in models/
4. Hyperparameter Tuning and Model Optimization, Complexity, Generalization, and Interpretation (5 Marks)
   • Done in model_tests.ipynb and saved to saved_models/
   • Hyperparameter tuning done through GridSearchCV (sparse search space due to computational constraints)
   • Complexity & generalization shown through outputs within notebook and dashboard
   • Interpretation given in INTERPRETATION.md
5. Overall ML Pipeline and Automation (5 Marks)
   • eda.ipynb (ETL Pipeline) -> model_tests.ipynb (train models) -> dashboard.py (interpret & visualize)
6. Visualization and Video Reporting (5 Marks)
   • Done in dashboard.py and Team07_Assignment2_Video.mp4
7. Code Demo and Viva (10 Marks - Individual Assessment)

PROBLEM

Classify persons into being high or low risk for having an emergency room admission.

Target

• HIGH_EMERGENCY_RISK: Positive (1) if >2 emergency room encounters logged, else Negative (0)

Features

• Age
• Medications
• Conditions
• Allergies
• Income
• Healthcare Coverage
• Procedure Cost

Models

• Decision Tree
• Neural Network (MLP)
• Support Vector Machine

Metrics

• Precision
• Accuracy
• Recall
• F1