ashfarhangi/AA-Forecast

Repository for the AA-Forecast paper ECML PKDD 2022

AA-Forecast: Anomaly-Aware Forecast for Extreme Events

Published in: Data Mining and Knowledge Discovery, 2023

Overview

Time series models are often impacted by extreme events and anomalies, which are prevalent in real-world datasets. Such models require careful probabilistic forecasts, vital in risk management for extreme events like hurricanes and pandemics. Our proposed framework, AA-Forecast, leverages the effects of anomalies to improve prediction accuracy during extreme events.

Key Features

• Automatic Anomaly Detection: The model extracts anomalies automatically and incorporates them through an attention mechanism to enhance forecast accuracy during extreme events.
• Dynamic Uncertainty Optimization: Employs an algorithm that reduces the uncertainty of forecasts in a nonlinear manner.
• Superior Performance: Demonstrates consistent superior accuracy with less uncertainty across different datasets compared to current prediction models.

Contributions

1. Anomaly Extraction: Introduces a novel decomposition method to extract anomalies.
2. Attention Mechanism: Leverages the extracted anomalies through a specialized attention mechanism to improve forecasting.
3. Dynamic Optimization: Reduces uncertainty in forecasts with a dynamic optimization approach.

Datasets

The framework is evaluated on the following datasets:

Hurricane Data: Time series data related to hurricane events. Pandemic Data: COVID-19 impact on various sectors. Synthetic Data: Generated datasets to test the model under controlled conditions.

Getting Started

Instructions on setting up your project locally or on a cloud platform. To get a local copy up and running follow these simple example steps.

Prerequisites

This is an example of how to list things you need to use the software and how to install them.

• Tensorflow 2.1.1
• Nvidia GPU

Datasets

Datasets are located in the data folder:

credit-card-sales-covid-19.csv electricity.csv tax-sales-hurricane.csv

Installation

1. Clone the repo.

   git clone https://github.com/0415070/AA-RNN.git
   

2. Install requirement packages.

   pip install -r requirements.txt
   

3. Run model.py after the dataset has been gathered. You can use make_data.py for this.

The citations for the paper, code, and data are as below:

@article{farhangi2022aa,
  title={AA-Forecast: Anomaly-Aware Forecast for Extreme Events},
  author={Farhangi, Ashkan and Bian, Jiang and Huang, Arthur and Xiong, Haoyi and Wang, Jun and Guo, Zhishan},
  journal={Data Mining and Knowledge Discovery},
  year={2023},
  volume={37},
  pages={1209-1229},
  doi={10.1007/s10618-023-00919-7}
}