SkyCast – Rainfall Prediction Web App

SkyCast is a cutting-edge web application designed to predict rainfall probabilities using advanced machine learning techniques. Tailored for tropical and subtropical coastal climates.

---

Key Features

• Intuitive User Experience
  A thoughtfully designed interface that prioritizes user engagement and ease of navigation, making weather predictions accessible to everyone.

• Advanced Machine Learning
  Utilizes an XGBoost classifier, trained on extensive real-world weather data, to deliver precise rainfall forecasts.

• Real-Time Predictions
  Instantly generates rainfall probabilities, categorizes risk levels, and provides actionable advice tailored to the forecast.

• Exploratory Data Analysis Tools
  Integrated utilities for visualizing and analyzing weather patterns, empowering users to understand trends and anomalies.

• Robust Input Validation
  Ensures data integrity with checks that prevent common input errors, enhancing the reliability of predictions.

---

Installation

To set up SkyCast, follow these steps:

git clone https://github.com/ayushsyntax/SkyCast.git
cd SkyCast
pip install -r requirements.txt
streamlit run app.py

The application will launch at http://localhost:8501.

---

How to Use

1. Enter the following details:

   • Date
   • Max Temp (°C) / Min Temp (°C)
   • Humidity (%) (using a slider)
   • Cloud Cover: Clear, Partly Cloudy, Cloudy
   • Wind Speed: Low, Moderate, High

2. Click “Predict Rainfall”
   The output will include:

   • Probability (% chance)
   • Risk Level: Low / Medium / High
   • Advice (e.g., “Carry an umbrella.”)

---

Model Insights

Feature Engineering

• Transforming the date into cyclical features to capture seasonal trends.
• Creating derived metrics such as temperature range and humidity combined with cloud cover.
• Implementing fallback defaults for missing data to maintain prediction accuracy.

Model Architecture

• Core Model: XGBoost Classifier
• Dataset: Sourced from Kaggle Playground S5E3
• Output: Provides a probability score along with a risk classification.

Postprocessing Techniques

• Establishing risk thresholds based on calibrated cutoffs.
• Mapping recommendations to specific risk outcomes for user guidance.

---

Exploratory Data Analysis Toolkit

Available in projectrain.py:

Library	Purpose
numpy	Efficient numerical operations
pandas	Data manipulation and analysis
matplotlib	Visualization of trends and patterns
seaborn	Advanced statistical graphics

---

System Overview

Component	Description
Model	XGBoost Classifier
Baseline	Logistic Regression
AUC-ROC	XGB: > 0.90 · LogReg: ~0.88
Framework	Streamlit
Python Version	3.x
Total Features	16 (including engineered inputs)

---

Directory Structure

SkyCast/
├── app.py              # Main application logic
├── projectrain.py      # EDA and preprocessing scripts
├── final_model.pkl     # Serialized machine learning model
├── requirements.txt    # List of dependencies
└── README.md           # Project documentation

---

Performance Overview

• Dataset: Comprised of 6 years of coastal climate data.
• Class Distribution: 75% rain / 25% no rain.
• Imbalance Handling: Addressed through class weighting.
• Performance Metrics:
  • XGBoost AUC-ROC: > 0.90
  • Logistic Regression AUC-ROC: ~0.88

---

Target Climate

SkyCast is specifically designed for:

• Regions: Southeast Asia, South India, coastal Australia.
• Weather Patterns: High humidity, cloud cycles, and moderate wind conditions.

---

Modeling Considerations

• Focused on a clean binary classification task.
• Key features influencing predictions include humidity, cloud cover, dew point, and temperature range.
• Seasonal signals are captured through cyclical time transformations.
• Utilizes stratified validation and class weights to ensure balanced predictions.
• Fully reproducible analysis available in projectrain.py.

---

Feedback and Contributions

We welcome your suggestions, feedback, and contributions to enhance SkyCast. If you encounter any issues or have ideas for new features, please feel free to reach out. You can open a GitHub issue or submit a pull request to contribute to the project.

---