Abstract
Despite rapid advances in automated text processing, many related tasks in transit and other transportation agencies are still performed manually. For example, incident management reports are often manually processed and subsequently stored in a standardized format for later use. The information contained in such reports can be valuable for many reasons: identification of issues with response actions, underlying causes of each incident, impacts on the system, etc. In this article, we develop a comprehensive, pragmatic automated framework for analyzing rail incident reports to support a wide range of applications and functions, depending on the constraints of the available data. The objectives are twofold: a) extract information that is required in the standard report forms (automation), and b) extract other useful content and insights from the unstructured text in the original report that would have otherwise been lost/ignored (knowledge discovery). The approach is demonstrated through a case study involving analysis of 23,728 records of general incidents in the London Underground (LU). The results show that it is possible to automatically extract delays, impacts on trains, mitigating strategies, underlying incident causes, and insights related to the potential actions and causes, as well as accurate classification of incidents into predefined categories.
Introduction
In transit management, detailed incident reports are essential for understanding system disruptions and improving operational resilience. However, the current manual processing of these reports by transit staff is inefficient and prone to "information leakage." Qualitative details regarding the root causes of delays or specific mitigation strategies are often lost when reports are condensed into rigid, standardized spreadsheets. This research addresses these challenges by developing a pragmatic, automated framework for analyzing rail incident reports, aiming to both automate standard data entry and discover deeper insights from unstructured text.
Methodology
The proposed framework utilizes a multi-tiered Natural Language Processing (NLP) pipeline designed to handle the complexities of transit data. For structured data extraction—such as Train IDs, station names, and delay durations—a Bidirectional Long Short-Term Memory (Bi-LSTM) model with a Conditional Random Field (CRF) layer is employed. This model is enhanced by BERT (Bidirectional Encoder Representations from Transformers) embeddings to handle domain-specific jargon and contextual nuances.
Beyond basic extraction, the system uses dependency parsing via the Stanford CoreNLP library to link impacts to specific entities. For knowledge discovery, Latent Dirichlet Allocation (LDA) is used for unsupervised topic modeling of incident causes. Automated classification of reports into categories like Fleet or Signals is handled by a ResNet-style 1D Convolutional Neural Network, while rule-based regular expressions capture conditional "lessons learned" logic that identifies how incidents might have been avoided.
Case Study and Results
The framework was validated using a real-world dataset from Transport for London (TfL) consisting of 23,728 incident records from the London Underground. The Bi-LSTM CRF model achieved high F1 scores, notably 0.94 for Train IDs and delay durations, and 0.88 for station names. The BERT-ResNet classifier reached an overall F1 score of 0.87, significantly outperforming traditional logistic regression baselines (F1 0.79).
Crucially, the system identified 434 instances where manual delay reporting was inaccurate or missing, demonstrating the system's ability to correct the human record. Furthermore, the LDA model successfully decomposed broad categories like "Staff" into 20 granular sub-topics, such as "Train operator not in position," providing a much higher resolution of operational issues than manual categorization.
Conclusion
This study demonstrates that deep learning and NLP can significantly enhance transit disruption management. By automating the extraction of both structured and qualitative data, agencies can build more resilient systems and preserve valuable institutional knowledge. Future research will focus on real-time integration to assist dispatchers and the analysis of external data sources like social media and passenger feedback.
Summary updated