Using an LSTM (Long Short-Term Memory) algorithm on synthetic time series data for unsupervised anomaly detection. It uses the Mean Squared Error as a measure of outliers in this type of data.

Diagram of original signal, smoothed signal and anomaly detection via MSE.

1. Install the Python dependencies (please use Python 3.6+).

pip install -r requirements.txt

2. Run the program (creates its own synthetic data, creates an LSTM model, trains it and predicts with it).

python run.py

That's it!

Some TODO:

• Keras (TF backend)
• Estimates on real-world data (e.g. this Kaggle rain dataset)
• PyTorch port

Other Approaches:

• Finite state automaton with state transitions for time series data segmentation (Ref)
• One-class NN (Code)
• TBD

1. https://github.com/aurotripathy/lstm-anomaly-detect
2. https://github.com/Vict0rSch/deep_learning/tree/master/keras/recurrent
3. http://simaaron.github.io/Estimating-rainfall-from-weather-radar-readings-using-recurrent-neural-networks/
4. https://github.com/khundman/telemanom/tree/no-labels

This work is partially based upon the Telemanom project with code and information found at https://github.com/khundman/telemanom.

If you use this work, please cite:

  title={Detecting Spacecraft Anomalies Using LSTMs and Nonparametric Dynamic Thresholding},
  author={Hundman, Kyle and Constantinou, Valentino and Laporte, Christopher and Colwell, Ian and Soderstrom, Tom},
  journal={arXiv preprint arXiv:1802.04431},
  year={2018}
}