Air Quality Analysis, Prediction and Visualization of Major Chinese Cities

This project presents a complete pipeline for air quality forecasting, from data analysis to model training and interactive visualization. Developed for the SDUST Pioneer Intelligent Computing 2025 competition, it is organized into three major parts:

• 📊 Data Analysis
• 🧠 LSTM Model Training
• 📈 Dash-based Web App

📁 Repository Structure

├── analysis/       # Data exploration and preprocessing scripts
├── model/          # LSTM model training and evaluation
├── app/            # Dash web app for PM2.5 prediction and visualization
├── README.md       # Project overview and instructions

1️⃣ Data Analysis

Located in the analysis/ folder.

• Performs exploratory data analysis (EDA) on historical air quality data.
• Generates plots and statistics for pollutants such as PM2.5, PM10, CO, SO₂, NO₂, and O₃.
• Handles data cleaning and preparation before model training.

Highlights:

• Outlier detection
• Missing value handling
• Feature correlation studies
• Important event analysis

2️⃣ LSTM Model Training

Located in the model/ folder.

• Implements a unified LSTM architecture using TensorFlow/Keras.
• Trains the model to predict PM2.5 levels using various pollutant features.
• Includes scaler serialization (using joblib) and model saving in both .h5 and .keras formats.
• Ready for hyperparameter tuning.

3️⃣ Interactive Dash App

Located in the app/ folder.

• Built with Plotly Dash for visualization and real-time inference.
• Users can select dates, locations, and view predicted vs. actual PM2.5 levels.
• Visualizes weather data from 2014 to 2025.

Features:

• Visualize weather data
• Interactive pollutant time series
• Model prediction interface
• Weather quality map
• User-friendly dashboard

🚀 Getting Started

1. Clone the repository:

   git clone https://github.com/alfa-echo-niner-ait/sdust-pilotcup-2025.git
   cd sdust-pilotcup-2025

2. Create & activate the virtual environment: Here's a quick guide for creating and activating a Python virtual environment (.venv) on both Windows and Linux:

   Create Virtual Environment

   python -m venv .venv

   (Use python3 instead of python on Linux if needed.)

   Activate Virtual Environment Windows (CMD/PowerShell)

   :: CMD
   .\.venv\Scripts\activate
   
   # PowerShell
   .\.venv\Scripts\Activate.ps1

   Linux/macOS (Bash/Zsh)

   source .venv/bin/activate

3. Install dependencies:

   • Each main folder (analysis/, model/, app/) contains its own requirements.txt.
   • Install as needed, for example:

     pip install -r analysis/requirements.txt
     pip install -r model/requirements.txt
     pip install -r app/requirements.txt

   • For users in China:

     pip install -r analysis/requirements.txt -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
     pip install -r model/requirements.txt -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
     pip install -r app/requirements.txt -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple

4. Run Data Analysis:

   • Use Jupyter notebooks in analysis/.

5. Train and test the Model:

   • Run and modify the Jupyther notebook in model/.

6. Launch the Web App:

   • Start the Dash app from app/:

     python app/app.py

📽️ App Preview

🧑‍💻 Authors

Team Name	NEXUS
Team Members	Ayub Ali Emon	[email protected]
	Phonesamay Phouthavong	[email protected]
	Yu Jinling	[email protected]
	Fu Wenting	[email protected]
Project Supervisor	Jiehan Zhou	[email protected]

🔗 Member of CogTwins Lab

🏫 Shandong University of Science and Technology

📄 License

This project is licensed under the MIT License.