Enhancing Safety-Critical Systems with Robust Ensemble Learning Algorithm

Abstract

Introduction: In the previous researches, Machine learning algorithms such as conventional ML algorithms have been used and have shown inability to provide the required level of accuracy and robustness for such applications, therefore Ensemble learning which combines multiple ML models has been used to improve the accuracy and robustness of ML algorithms. The development of safety critical systems relies on stringent safety methodologies, designs, and analyses to prevent hazards at the time of failure. Several studies have proposed different ML techniques, such as Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest (RF), and Deep Neural Networks (DNN) to predict soft-error prior to the occurrence. Hence we here discuss how we will use the ensemble algorithm to enhance prediction of soft-errors in critical systems.

Objectives: implementation of robust algorithm to enhance prediction of soft errors in safety critical systems.

Methods: the methodology involves applying XGBoost and LightGBM learning algorithms to predict potential errors in pacemakers. This approach is used to overcome the limitations of individual ML algorithms. The selection of a specific ensemble algorithm considers factors such as data characteristics, interpretability, performance, feature engineering, the ensemble method itself, validation, and regulatory considerations to optimize the prediction of soft errors in pacemakers

Results: the robust ensemble algorithms achieved the objective of enhancing the safety critical systems through prediction of soft errors there achieving the prediction accuracies of 0.98 and 0.99 for light gradient boosting algorithm and extreme gradient boosting respectively.

Conclusions:  The implemented robust ensemble algorithm selected for soft error analysis and mitigation deploys Boosting techniques. The algorithm is capable of handling high-dimensional and noisy data hence able to generate accurate predictions