Answers are grounded in this paper and cite their sources.
Lasers are widely regarded as the most critical components in optical communication systems. Their reliability has a significant impact on the system's availability and performance. Conventionally, the methods adopted for laser reliability estimation are based on extrapolation to experimental reliability data derived from accelerated aging tests, and could lead to either overestimation or underestimation of actual laser lifetime. Alternatively, machine learning-based approaches have shown great promise in improving laser reliability and outperforming traditional methods. Nonetheless, one major barrier to the adoption of ML methods is their lack of interpretability as they operate as black-box methods. To address this issue, we propose an interpretable ML model for laser lifetime prediction. A Shapley additive explanations (SHAP) method is used to explain the predictions made by the ML approach by quantifying the relevance of inputs, clarifying their effects on each individual estimation, and emphasizing their interaction. The proposed approach is validated using synthetic reliability data. The results show that the ML model achieves a good predictive performance (yielding a root mean square error of 0.37 years). The interpretability analysis shows that the slope efficiency and the power are the top two most significant features impacting the laser lifetime predictions made by the ML model. The predictive accuracy of the ML model can be further improved slightly after re-examination of features.
No takes yet. Share an insight, caveat, or question.
Abdelli et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: