Next-Gen Air Quality Forecasting using Multi-modal Deep Learning (Images + sensor data) with a built-in research-grade Explainable AI (XAI) suite.
MultiModal AQI is not just a predictor; it's an interpreter. By fusing high-resolution environmental imagery (CNN) with 24-hour pollutant telemetry (LSTM), it provides a holistic view of urban air quality.
Go beyond the number. Understand the Why behind every prediction:
- Grad-CAM: Visual heatmaps pinpointing exactly which parts of an image (e.g., traffic, haze, skyline) the AI is "looking" at.
- SHAP Analysis: Calculating the game-theoretic contribution of each pollutant (PM2.5, NO2, etc.) to the final AQI score.
- LIME Occlusion: Brute-force sensitivity analysis by systematically masking image regions to measure prediction stability.
- Counterfactuals: "What-if" simulations showing how a 10% drop in PM2.5 would shift the AQI category.
*Modern SaaS-style interface with real-time stats and multi-modal forecasting.*
Detailed Grad-CAM heatmap analysis and regional contribution breakdown.
- 🎭 Multi-modal Fusion: Seamlessly combines CNN-extracted visual features with LSTM-processed time-series data.
- 📊 SaaS Analytics: Real-time session metrics, average confidence tracking, and anomaly detection.
- ⚙️ Model Configuration: Live view of the underlying CNN+LSTM+Attention architecture.
- 🔀 Simulator: Modify pollutant sliders to see real-time "What-If" prediction shifts.
- 📱 Responsive Design: Fully optimized for mobile and desktop using a premium Glassmorphism theme.
Frontend:
- React 18 + Vite (Next-gen bundling)
- TypeScript (Type safety)
- Tailwind CSS (Premium styling)
- Lucide React (Vector iconography)
- Recharts (Visual analytics)
- Framer Motion (Smooth transitions)
Backend:
- FastAPI (High-performance API)
- TensorFlow/Keras (Deep learning core)
- OpenCV (Image processing & Heatmap generation)
- SHAP & LIME (Explainability logic)
- Pandas/NumPy (Data processing)
- Python 3.9+
- Node.js 18+
- Git
cd backend
python -m venv .venv
# Activate: source .venv/bin/activate (Linux/Mac) or .venv\Scripts\activate (Windows)
pip install -r requirements.txt
uvicorn app:app --reload --port 8000cd frontend
npm install
npm run dev├── backend/ # FastAPI Server & XAI Logic
│ ├── routes/ # API Endpoints
│ ├── training/ # Model Architecture & Training Scripts
│ └── utils/ # Heatmap & Preprocessing utilities
├── frontend/ # React + Vite Dashboard
│ ├── src/ # Components & Logic
│ └── public/ # Static Assets
└── model/ # Trained weights (.keras files)
Contributions are welcome! Please open an issue or submit a pull request for any improvements or bug fixes.
This project is licensed under the MIT License.
Developed by Sameer Senapati 🚀