Unsupervised Outlier Detection in IOT Using Deep VAE.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Internet of Things (IoT) systems generate vast amounts of data, often containing outliers due to device failures or cyberattacks.
• Existing outlier detection methods typically require labeled data, which is scarce and costly to obtain in the IoT domain.
• The increasing data volume in IoT necessitates efficient, unsupervised anomaly detection techniques for real-time monitoring and security.

Purpose of the Study:

• To propose an unsupervised outlier detection technique for IoT data using a deep Variational Auto-Encoder (VAE).
• To leverage the VAE's reconstruction capabilities and latent variable representation for anomaly identification.
• To address the challenge of limited labeled data in IoT environments.

Main Methods:

• Standardization of input IoT data.
• Application of a deep Variational Auto-Encoder (VAE) to learn a low-dimensional latent representation and reconstruct input data.
• Calculation of reconstruction error between original and reconstructed data as an outlier score.

Main Results:

• The unsupervised VAE model achieved high performance on the Statlog (Landsat Satellite) dataset.
• The model demonstrated comparable results to state-of-the-art outlier detection schemes.
• Achieved a precision of approximately 90% and an F1 score of 79% using only normal data for training.

Conclusions:

• The proposed unsupervised VAE-based method is effective for outlier detection in IoT data.
• This approach overcomes the limitations of labeled data dependency in traditional methods.
• The technique offers a promising solution for anomaly detection in large-scale, unlabeled IoT datasets.