FloodML

FloodML is a web application that leverages Machine Learning to predict floods based on weather and historical data.

Overview

Getting Started

• Clone the project
• Install dependencies
• Activate virtual environment
• Run pip install -r requirements.txt
• Run python app.py

Inspiration

Floods, exacerbated by climate change, pose a growing threat worldwide. To address this, we created FloodML—an interactive web app for predicting and visualizing floods.

Core Components

1. Plots

• Flood Prediction Plot (Red dots for predicted flood locations)
• Precipitation Plot (Bubbles indicate precipitation volume)
• Damage Analysis Plot (Bubble size represents estimated monetary damage)

2. Heatmaps

• Damage Analysis Heatmap (Colors indicate predicted monetary damage)
• Precipitation Heatmap (Dark red areas signify higher precipitation)
• Flood Prediction Heatmap (Darker red spots indicate likely flood locations)

3. Satellite Images

• Displays precipitation volume over Indian cities
• Uses NASA's Global Precipitation Measurement Project data

4. Predict Page

• Real-time weather forecast and flood prediction for any city
• Includes temperature, humidity, cloud cover, wind speed, and precipitation

Development Process

The Dataset

• Scraped floodlist.com using Beautiful Soup 4
• Utilized Visual Crossing weather API for historic weather data
• Applied data augmentation techniques for model diversity

ML Model

• Built on the sci-kit learn library
• Explored various models; Random Forest Classifier achieved 98.71% accuracy
• Saved model using pickle

Data Visualization

• Integrated major Indian cities' data with weather factors
• Utilized Plotly chart studio for diverse map visualizations

Front-end and Hosting

• Developed with Flask framework
• Hosted on Heroku

Challenges

• Limited data availability for floods in India
• Plotly integration complexities
• Git merge conflicts due to encoding and version disparities

Achievements

• Created a robust dataset for accurate flood predictions
• Implemented a machine learning model with over 98% accuracy
• Successfully integrated data augmentation and visualization techniques

Learnings

• Enhanced skills in web scraping, data mining, and Plotly
• Expanded proficiency in machine learning models

Future Plans

• Expand coverage to cities worldwide
• Implement image classification for flood detection using satellite data

FloodML aims to aid people and governments in flood preparation, potentially saving lives and livelihoods. Visit FloodML to explore the tool.