Predictive maintenance using tree-based classification techniques: A case of railway switches

摘要: Abstract With growing service demands, rapid deterioration due to extensive usage, and limited maintenance budget cuts, the railway infrastructure is in a critical state require continuous maintenance. The managers have come up with smart decisions order improve assets’ condition, spend optimal cost keep network available. Currently, lack tools decision support models that could assist them taking (un) planned effectively efficiently. Recently, many literature studies proposed employ machine learning techniques estimate performance of an asset, predict need, possible failure modes, such similar aspects advance. Most these utilised additional data collection measures record behaviour. Though useful for experimentation, it expensive impractical mount monitoring devices on multiple assets across network. Therefore, objective this study develop predictive utilise existing from agency yield interpretable results. We propose leverage tree-based classification activity type trigger’s status switches. Using in-use business process, based tree, random forest, gradient boosted trees are developed. Moreover, facilitate interpretability, we provided detail explanation models’ predictions by features importance analysis instance level details. Our solution approach development their results wider applicability can be used other asset types different (maintenance) planning scenarios.