Anomaly-Detection

Anomaly detection methods and implementation.

Preprocessing

• utils.py - Trajectory matrix/ rolling window representation of time series data.

Algorithms

• LSTM Autoencoder
  • Long short-term memory (LSTM) is a recurrent neural network (RNN) architecture, used to process entire sequences of data.
  • An autoencoder learns a representation for a set of data in an unsupervised manner by training the network to ignore signal “noise”.
  • The reconstruction error serves as an indicator to the extent of a data instance being an anomaly.

Datasets

• Numenta Anomaly Benchmark
  • machine_temperature_system_failure.csv
    • Temperature sensor data of an internal component of a large, industrial machine.
    • The first anomaly is a planned shutdown of the machine.
    • The second anomaly is difficult to detect and directly led to the third anomaly, a catastrophic failure of the machine.