ML API – Crowd & Panic Intelligence Service

A production-ready Machine Learning API that serves real-time predictions for crowd load, panic indicators, and risk signals.
Designed to seamlessly integrate with websites, dashboards, and surveillance systems.

---

Purpose

This API acts as the intelligence layer for crowd & panic detection systems.
It exposes ML models via REST endpoints to:

• Predict crowd footfall
• Consume real-time inputs (camera data, weather, calendar, events)
• Continuously store actual counts for retraining
• Serve predictions with low latency

---

Core Capabilities

• Ensemble ML inference (CatBoost + LightGBM / XGBoost + LSTM)
• Feature-engineered predictions (calendar, weather, festivals)
• RESTful API using FastAPI
• MongoDB for real-time data storage
• CORS-enabled for frontend integration
• Scalable & modular architecture

---

Architecture Overview

Client / Website / Dashboard | v ML API (FastAPI) | ├── Feature Engineering Layer ├── ML Ensemble Models ├── Panic / Risk Logic | MongoDB (Actuals + Logs)

---

Tech Stack

• Python
• FastAPI
• Uvicorn
• Scikit-learn
• CatBoost
• LightGBM / XGBoost
• TensorFlow (LSTM)
• MongoDB
• Joblib

---

Project Structure

ml_api/ ├── app.py ├── test_clean_encoded_2.csv ├── models/ │ ├── Blended_model.pkl │ └── scalers.pkl ├── requirements.txt └── README.md

---

Installation

git clone https://github.com/jinay-k-jain/ml_api.git
cd ml_api
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Run the API

uvicorn app:app --host 0.0.0.0 --port 8000

ML Details (High Level) -LSTM → captures temporal crowd patterns -Tree Models → capture nonlinear interactions -Meta Model → combines all predictions -Feature Engineering includes: -Day / month cyclic encoding -Festival impact -Weather influence -Lag & rolling statistics

Data Storage MongoDB Stores daily actual footfall Stores predictions vs reality Enables continuous retraining

👤 Author Jinay Jain B.Tech, IIT (ISM) Dhanbad Smart India Hackathon Winner