Flood Risk Prediction Model

A machine learning solution for predicting flood probability based on environmental and geographical factors. This project implements multiple regression models and compares their performance on a large-scale dataset.

Overview

The project analyzes 20+ environmental features to predict flood probability, including monsoon intensity, deforestation, infrastructure quality, and climate change indicators. The dataset contains over 1.1 million training samples from the Kaggle Playground Series competition.

Models

Three regression models are implemented and evaluated:

• Linear Regression: R² = 0.828
• Gradient Boosting Regressor: R² = 0.619
• LARS: R² = 0.001

Linear Regression achieves the best performance with MAE of 0.329.

Dataset

Features include environmental and geographical factors:

• Monsoon Intensity, Topography, River Management
• Deforestation, Urbanization, Climate Change
• Infrastructure and population metrics
• Historical disaster preparedness data

Dataset source: Kaggle Playground Series S4E5

Project Structure

flood_prediction.ipynb          # Model training and evaluation
data/flood/
├── train.csv                   # Training dataset (1,117,957 samples)
├── test.csv                    # Test dataset (745,305 samples)
└── sample_submission.csv       # Submission format

Requirements

• Python 3.11+
• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn

Usage

1. Place dataset files in data/flood/ directory
2. Open and run flood_prediction.ipynb in Jupyter or VS Code
3. The notebook performs EDA, feature engineering, and model training
4. Generated predictions can be formatted for competition submission

Data Processing

• Outlier detection using IQR method
• Feature standardization using StandardScaler
• No missing values detected in the dataset
• 845,886 samples retained after outlier removal