A classification tree-based system for multi-sensor train approach detection Personnel safety is an integral element of any railroad operation. Rail tracks need to be regularly maintained, which often requires rail workers to be physically present on possibly live tracks along with their equipment. This results in hazardous work conditions for the workers. To ensure worker safety a reliable system for detecting oncoming trains and alerting the workers, while giving them sufficient time to disengage from the worksite, is essential. In this paper, we present a multiple sensor based system that integrates the sensor elements with a signal processing unit for this purpose. The processing unit consists of a signal conditioning unit, a data processing unit and a machine learning framework. The conditioning unit prepares the acquired signals for later operations. The data processing unit extracts fingerprints from the reported signals that are later used by the machine learning framework as training and testing samples. The machine framework is a binary tree that classifies the event under investigation as presence or absence of a train on the track under observation. We show that the system is very accurate and can alert the workers under five seconds after the arrival of the train at the test site.