GSX: Gumbel-Sigmoid eXplanator for Explainable Fault Prediction

TensorFlow implementation of GSX (Gumbel-Sigmoid eXplanator) for instance-wise feature selection and explainable fault prediction on IoT vibration time-series data.

---

Description

This repository contains the implementation of the GSX framework proposed for fault prediction using IoT sensor data with instance-wise explanations.

The core idea is to train a predictive model (e.g., LSTM/CNN-based) and then learn a differentiable selector that highlights which time steps (features) are most responsible for a prediction. GSX uses a Gumbel-Sigmoid selection mechanism to support gradient-based optimization while enabling sparse, instance-specific explanations.

The repository also includes comparison methods such as L2X and INVASE, together with multiple deep learning baselines for prediction.

---

Paper

This repository accompanies the following paper:

T. Mansouri and S. Vadera,
“A Deep Explainable Model for Fault Prediction Using IoT Sensors”,
IEEE Access, vol. 10, pp. 66933-66942, 2022.
DOI: 10.1109/ACCESS.2022.3184693

---

Citation

If you use this repository in your research, please cite:

BibTeX

@article{mansouri2022deepExplainableFaultPrediction,
  author  = {Taha Mansouri and Sunil Vadera},
  title   = {A Deep Explainable Model for Fault Prediction Using IoT Sensors},
  journal = {IEEE Access},
  volume  = {10},
  pages   = {66933--66942},
  year    = {2022},
  doi     = {10.1109/ACCESS.2022.3184693}
}