PredictaCare — Machine Learning for Healthcare Optimization

PredictaCare is a Philly Codefest project (Drexel’s 11th annual “Leveraging AI for the Common Good”) that predicts healthcare appointment durations to help clinics optimize scheduling and reduce patient wait time.

April 2024 – June 2024 • Team: Jeff Torr, Eli Young, Tony Stanell

What it does

• Analyzes a short free‑text appointment description plus demographics (age, gender, prior visit) and predicts the estimated visit length (in minutes).
• Supports a simple web UI and an HTTP endpoint for programmatic use.

Tech stack

• Python (scikit‑learn, NLTK, NumPy, SciPy, pandas)
• Node.js + Express for the web/API server
• Static HTML/CSS/JS frontend
• Dockerized runtime for reproducible setup
• License: GPL‑3.0 (see LICENSE)

Repository structure

• nodejs-docker-web-app/: Main app (Node.js server, frontend, Dockerfile, Python model code and CSV dataset)
• BackEndTerminalTests/: Historic/experimental terminal scripts and models used during exploration

---

Getting started

You can run PredictaCare either with Docker (recommended) or locally.

Option A: Docker

1. Open a terminal and navigate to the app folder:

cd nodejs-docker-web-app

2. Build the image:

docker build -f dockerfile -t predictacare:latest .

3. Run the container:

docker run --rm -p 8080:8080 predictacare:latest

4. Open the UI at http://localhost:8080

Option B: Local (Node.js + Python)

Prerequisites: Node 18+, Python 3.9+.

1. Install Node dependencies:

cd nodejs-docker-web-app
npm install

2. Install Python dependencies:

pip install -r requirements.txt

3. On Windows, ensure the app uses the right Python command (set once per shell):

set PYTHON_EXECUTABLE=python

On macOS/Linux (optional; defaults to python3):

export PYTHON_EXECUTABLE=python3

4. Start the server:

npm start

5. Open http://localhost:8080

---

API

POST /run-python

• Request body (JSON):

{
  "descriptionField": "sore throat and fever",
  "ageField": 37,
  "genderField": "male",   // "male" | "female"
  "familiarityField": "yes" // "yes" | "no"
}

• Response (JSON):

{ "output": "23.418273" }

Example with curl:

curl -X POST http://localhost:8080/run-python \
  -H "Content-Type: application/json" \
  -d '{
    "descriptionField": "sore throat and fever",
    "ageField": 37,
    "genderField": "male",
    "familiarityField": "yes"
  }'

Notes:

• output is the predicted appointment length in minutes (stringified).
• The Node server shells out to a Python script which loads/trains a model and returns a prediction.

---

Model and data

• Dataset: EstimateHealthcareAppointmentLengthGivenX-Sheet2.csv
• Features:
  • Text: appointment description → TF‑IDF (1–2 grams)
  • Numeric: age, gender (0/1), familiarity (0/1)
• Model: scikit‑learn Linear Regression trained on TF‑IDF + numeric features (concatenated via SciPy sparse hstack)
• Preprocessing: tokenization, lowercasing, lemmatization, English stopword removal (NLTK)

Training/prediction entry point: nodejs-docker-web-app/SkLearnLinearRegression.py

---

Development notes

• The Docker image installs NLTK data at build time to avoid first‑run downloads.
• On Windows, set PYTHON_EXECUTABLE=python before npm start if python3 is not available.
• This repository includes earlier terminal‑only experiments in BackEndTerminalTests/ for reference.

---

Safety and limitations

PredictaCare is a research/demo project. It is not a medical device and must not be used for diagnosis or treatment decisions.

---

Acknowledgments

Built and showcased at Drexel’s 11th annual Philly Codefest: “Leveraging AI for the Common Good.”