Scholarlink

<p>Diligent railway track irregularities maintenance scheduling is important for a safe rail traffic operation. Machine learning classifiers based on vehicle response have been shown to be an effective approach for track irregularities assessment. However, when simultaneously assessing several types of track irregularities within a given track section, loss of interpretation becomes a prevalent issue as the track irregularities assessment and the accompanying machine learning classification become more complicated. The present work examines the use of machine learning classification for combined track irregularities assessment and the subsequent result interpretation using Shapley Additive Explanation (SHAP) and Accumulated Local Effects (ALE). Testing results of the trained classification models show a high accuracy value, i.e., higher than 92%, with a low sensitivity against change in operational parameters. This indicates the suitability of this technique for track irregularities assessment. Furthermore, the interpretation analyses demonstrate a favourable potential in interpreting the outcomes of the track irregularities classification for specific sections. In particular, information from SHAP and ALE can be useful for identifying the threshold for acceptable acceleration levels. This feature is especially valuable for exploring the root causes of irregularities within a given track section. The interpretability not only enhances the ability to diagnose and address specific track irregularities but also underscores the potential for data-driven and data-informed approaches in the domain of railway track assessment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Interpretable machine learning classification for vehicle response-based railway track irregularities assessment

  • 期刊论文

  • 发表日期：

    2026年1月19日

查看全文
  • Prasidya Wikaranadhi,
  • Mats Berg,
  • Rohan Kulkarni,
  • Andi Isra Mahyuddin,
  • Pramudita Satria Palar,
  • Yunendar Aryo Handoko

摘要

Diligent railway track irregularities maintenance scheduling is important for a safe rail traffic operation. Machine learning classifiers based on vehicle response have been shown to be an effective approach for track irregularities assessment. However, when simultaneously assessing several types of track irregularities within a given track section, loss of interpretation becomes a prevalent issue as the track irregularities assessment and the accompanying machine learning classification become more complicated. The present work examines the use of machine learning classification for combined track irregularities assessment and the subsequent result interpretation using Shapley Additive Explanation (SHAP) and Accumulated Local Effects (ALE). Testing results of the trained classification models show a high accuracy value, i.e., higher than 92%, with a low sensitivity against change in operational parameters. This indicates the suitability of this technique for track irregularities assessment. Furthermore, the interpretation analyses demonstrate a favourable potential in interpreting the outcomes of the track irregularities classification for specific sections. In particular, information from SHAP and ALE can be useful for identifying the threshold for acceptable acceleration levels. This feature is especially valuable for exploring the root causes of irregularities within a given track section. The interpretability not only enhances the ability to diagnose and address specific track irregularities but also underscores the potential for data-driven and data-informed approaches in the domain of railway track assessment.