Autoencoder-LSTM Algorithm for Anomaly Detection
Hamid Akbarian, Imad Mahgoub, André Williams
Abstract
As the field of Artificial Intelligence (AI) continues to expand, AI-driven anomaly detection algorithms become paramount for operators to issue corrective actions, preventing disasters and reducing unnecessary costs. Historically, AI utilized deterministic rule-based techniques for anomaly detection. Today advances in AI have enabled more sophisticated algorithms. This paper proposes an Autoencoder Long Short-Term Memory (AE-LSTM) algorithm to improve anomaly detection. We evaluate and compare the efficacy of AE-LSTM against the benchmark Deep Neural Network Long Short-Term Memory (DNN-LSTM) algorithm. Evaluation metrics include false positives (FP), false negatives (FN), algorithm execution/run time, and F1-score. Autoencoder (AE) base architecture has been chosen to leverage its dimension reduction capabilities for relevant feature extraction. Our proposed scheme results show a considerable improvement from DNN-LSTM for anomaly detection which applies directly to space application and operation and was evaluated based on real-world datasets.