Hospital Crowd Prediction

Recommendation System

Official documentation for the **AI-powered system** that forecasts hospital traffic, optimizes patient visits, and enhances operational efficiency using machine learning.

Project By: Group 2

Devadath N.P

Project Leader

Abhirath S Madhav

Hanna Fathima TK

Fathimath Faheema

Krishnapriya

Ananya Sujay

1. Project Overview

The **Hospital Queue Optimizer** is an intelligent web application designed to forecast crowd levels in a hospital at any given day and time. By leveraging a machine learning model, it provides patients with data-driven recommendations for the optimal time to visit, minimizing wait times and improving patient experience.

Primary Goal

To predict hospital crowdedness and suggest the least busy time slot for a visit on a selected day, enhancing resource management and patient flow.

Core Technology

Built with Streamlit for the interactive UI and a Random Forest Classifier for the prediction engine, trained on synthetic hospital traffic data.

2. Prerequisites & Requirements

A **Python 3.x environment** is required. The project relies on several key data science and web development libraries, which are listed below.

  • streamlit — For building the interactive web application GUI.
  • numpy & pandas — For efficient data manipulation and handling.
  • scikit-learn — For the machine learning model (RandomForestClassifier, StandardScaler).
  • matplotlib & seaborn — For generating performance visualizations like the Confusion Matrix.
  • datetime — For handling date and time logic.

3. Setup Instructions

3.1. Clone the Repository

First, get the project source code by cloning the official GitHub repository.

# Clone the repository
git clone https://github.com/HospitalCrowd/Crowdprediction-AI.git

# Navigate into the project directory
cd Crowdprediction-AI

3.2. Install Dependencies

Install all required libraries using pip. It's recommended to do this within a virtual environment.

# Install the necessary libraries
pip install streamlit numpy pandas scikit-learn matplotlib seaborn

4. Running the Application

With the dependencies installed, you can start the application using the Streamlit CLI. The app will automatically generate the dataset and train the model on its first run.

(Assuming the main script is named app.py) 

# Execute the Streamlit application
streamlit run app.py

This command launches a local web server and opens the application in your default browser at http://localhost:8501.

5. Using the Streamlit Interface

User Inputs & Controls

The interface is designed for simplicity. All user controls are located in the sidebar:

  1. Select a Date: Choose your desired day using the calendar widget.
  2. Select a Time: Pick a specific time slot for the prediction (e.g., 10:00 AM).
  3. Predict & Suggest: Click this primary button to trigger the analysis.

Prediction & Recommendation

After clicking the button, the app instantly provides two key pieces of information:

  • Crowd Status: A clear indication of whether the hospital is predicted to be Crowded or Not Crowded at your selected time.
  • Best Time to Visit: A recommendation for the optimal 30-minute slot on the chosen day, identified by simulating all time slots to find the one with the lowest predicted crowd level.
Application Screenshot

6. Technical & Performance Details

The application provides transparency into the machine learning model's performance and architecture.

Model Type

Random Forest Classifier

Chosen for high accuracy, speed, and robustness against overfitting on structured data.

Model Accuracy

~90-92%

A high performance metric displayed directly in the user interface for full transparency.

Data Size

~15,000 Samples

Each data point represents aggregated crowd data for a 30-minute time slot.

Input Features

3 Core Features

Features used for prediction: hour, day_of_week, is_weekend.

Training Split

70% / 30%

The dataset is split 70% for model training and 30% for validation/testing.

Data Caching

@st.cache_data

Streamlit's caching decorator is used to optimize performance by loading the dataset and trained model only once.

Confusion Matrix

The application dynamically displays a confusion matrix, visually representing the model's performance by showing the counts of correct (True Positives/Negatives) versus incorrect (False Positives/Negatives) predictions, offering clear model diagnostics.

Classification Report

A detailed report is also generated, providing key classification metrics like **precision**, **recall**, and the **F1-score** for each class (Crowded vs. Not Crowded), ensuring a complete understanding of the model's predictive capabilities.