Unsupervised novelty detection for time series using a deep learning approach.

Area of Science:

• Computer Science
• Artificial Intelligence
• Internet of Things

Background:

• Smart Home Systems (SHS) generate vast data, necessitating intelligent anomaly detection.
• Existing methods often struggle with unlabeled data and distinguishing anomaly types.
• Novelty anomaly detection is crucial for maintaining SHS integrity and performance.

Purpose of the Study:

• To introduce DeepMaly, a novel unsupervised method for novelty anomaly detection in SHS.
• To enable effective anomaly identification in unlabeled time series data.
• To provide a practical tool for SHS developers to enhance system intelligence.

Main Methods:

• Utilized a combination of Long Short-Term Memory (LSTM) and Deep Convolutional Neural Network (DCNN).
• Employed unsupervised learning on unlabeled pristine features from time series data.
• Developed a data prediction and classification process for normal vs. abnormal data.

Main Results:

• DeepMaly successfully distinguished between seasonal and actual anomalies in an unsupervised manner.
• The method demonstrated prowess in novelty detection on benchmark datasets.
• Achieved real-time anomaly identification capabilities for SHS.

Conclusions:

• DeepMaly offers a practical solution for anomaly detection in unlabeled SHS datasets.
• The unsupervised approach reduces the need for extensive data labeling.
• Enhances the security and reliability of Smart Homes and IoT environments.