This repository is my PhD first year work.
The abstract of my work is as followed:
We propose a new two-level automatic feature extraction based unsupervised anomaly detection method for IIoT. It possesses two levels of feature extraction: 1) Autoencoder-based feature extraction and 2) SHAP-based explainable feature selection. We show that repeatedly doing feature extraction by autoencoder can get a higher accuracy for anomaly detection, and with the important features selected by SHAP explainer, the rationale of why an anomaly detection decision was made is given, enhancing the trustworthiness of the detection as well as further improving the accuracy of anomaly detection. Wafer real world IIoT datasets comprising with high-dimensional features are used to validate the effectiveness of the proposed unsupervised anomaly detection model. Extensive experiments show that our approach can get a higher accuracy for anomaly detection and enhance the trustworthiness of the detection.
Reference:
[1] Dataset: Detecting Anomalies in Wafer Manufacturing https://www.kaggle.com/arbazkhan971/anomaly-detection