Multi-Step Air Quality Forecasting

The "Multi-Step Air Quality Forecasting" project is designed to predict air quality levels using a systematic pipeline with four key components.

Overview

• Framework Used: Keras
• Components:
  1. Data Preprocessing
  2. Time-Series Decomposition
  3. Temporal Feature Extraction
  4. Prediction Layer

Components

Data Preprocessing

This component involves preparing the raw air quality data for analysis. It includes data normalization and transforming the data into a sliding window format to ensure compatibility with subsequent pipeline steps.

Time-Series Decomposition

Here, the time-series data is decomposed into trend and seasonality components. This decomposition aids in understanding underlying patterns and trends, facilitating more accurate forecasts.

Temporal Feature Extraction

Two LSTMs are utilized - one for the seasonal component and another for the trend component. Alternatively, an LSTM auto-encoder network is employed for seasonal and trend components in other architectures.

Prediction Layer

Extracted features and learned temporal dependencies are combined to predict air quality for future time steps. The prediction layer comprises a simple feed-forward neural network, which maps the features to predicted air quality values.

Integrating these components into a cohesive pipeline aims to develop accurate forecasts of air quality levels, enhancing decision-making and planning in areas affected by air pollution.

Environments

To set up the required environments, use the following commands:

pip install -U pip
pip install -r requirements.txt