This project aims towards obtaining reliability and robustness estimate for a machine learning (ML) component. In particular, we focus on time series application. The repository contains the recent work, function and results. Further, Jupyter notebooks exemplifying the work are also provided.
The statistical distance dissimilarity is used to estimate reliability and robustnessThe methodology is demonstrated on the following applications:
- Stock price prediction (Uni-variate time series)
- Machine failure prediction (multi-variate time series) [In progress]
To get started, simply create and activate the Anaconda environment with a name "stock" from the environment yml file provided.
conda env create -f env.yml
conda activate stock
The details examples can be found in the respective folders in UnivariateTimeSeries
.
- Akshatha Ambekar (Fraunhofer Institute for Experimental Software Engineering)
- Mohammed Naveed Akram (Fraunhofer Institute for Experimental Software Engineering) - naveed.akram@iese.fraunhofer.de
- Ioannis Sorokos (Fraunhofer Institute for Experimental Software Engineering) - ioannis.sorokos@iese.fraunhofer.de
- Koorosh Aslansefat (University of Hull) - k.aslansefat@hull.ac.uk
Akram, M. N., Ambekar, A., Sorokos, I., Aslansefat, K., & Schneider, D. StaDRe and StaDRo: Reliability and Robustness Estimation of ML-based Forecasting using Statistical Distance Measures.In International Conference on Computer Safety, Reliability, and Security SAFECOMP 2022. [Pre-print PDF]
@inproceedings{Akram2022Stadre,
author = {{Akram}, Mohammed Naveed and {Ambekar}, Akshatha and
{Sorokos}, Ioannis and {Aslansefat}, Koorosh and
{Schneider}, Daniel},
title = "{StaDRe and StaDRo: Reliability and Robustness
Estimation of ML-based Forecasting using Statistical
Distance Measures}",
booktitle = {International Conference on Computer Safety
Reliability, and Security},
year = {2022},
pages = {-},
organization = {Springer},
}
This work was partly supported by the Building Trust in Ecosystems and Ecosystem Component (BIECO) Horizon 2020 Project under Grant Agreement 952702.
This framework is available under the MIT License.