UNSUPERVISED MACHINE LEARNING FOR MANAGING SAFETY ACCIDENTS IN RAILWAY STATIONS

Abstract

Railroad operations must be reliable, easily 
accessible, well-maintained, and safe 
(RAMS) in order to move both passengers 
and freight. An important everyday safety 
issue in many metropolitan areas is the 
danger of accidents at train stations. In 
addition to casualties, public worry, and 
financial losses, accidents wreak havoc on 
market reputation. Stations like this are 
feeling the heat from increased demand, 
which is putting a strain on infrastructure 
and making safety a top administrative 
priority. The use of unsupervised topic 
modeling to better understand the 
contributors to these severe incidents is 
advised for the purpose of analyzing them 
and using technology, such as AI 
approaches, to increase safety. This study 
aims to optimize Latent Dirichlet Allocation 
(LDA) using textual data collected from 
RSSB, which includes 1000 accidents that 
occurred at UK railway stations, with the 
purpose of reducing mortality accidents at 
these locations. To improve station safety 
and risk management, this study details the 
use of a machine learning topic approach to 
systematic detect accident characteristics, 
and it offers advanced analysis. The research 
assesses the effectiveness of text mining 
from accident records in obtaining 
information, lessons learned, and a 
comprehensive understanding of the risks 
associated with evaluating accidents that 
result in deaths on a large and long-term 
scale. Predictive accuracy for important 
accident data, including hotspots at railway 
stations and underlying causes, is shown by 
this Intelligent Text Analysis. In addition, 
the advancement of big data analytics allows 
for a better understanding of the nature of 
accidents compared to restricted domain 
analysis of accident reports, which was 
previously impossible without a large 
quantity of safety history. A new age of 
useful and widespread artificial intelligence 
applications in railway sector safety and 
other domains for safety applications has 
dawned, made possible by this technology's 
high level of accuracy.