NFPA Data Lab

NFPA has been conducting mission-driven, data-oriented R&D since the early 20th century to advance the cause of fire and electrical safety.  To continue that effort in a digital and collaborative world, NFPA is committed to developing a suite of open-source tools, code bases, research workflows, and scripts for people to freely use, develop, and adapt to solve their own data challenges. 

These libraries along with documentation on how to use them can be found at NFPA’s Github page and are provided under permissive open-source licenses to encourage continued and collaborative innovation between NFPA, our stakeholders, and the broader data analytics community. In the spirit of open-source software, please note that all these projects should be considered continually “under development” and should viewed as collaborative, experimental tools and not necessarily production-ready platforms.

Recent open-source initiatives:
  • Zabaan: A prototype machine translation service developed by data scientists in NFPA’s Global division in collaboration with WPI GQP Program. The system aims to use a hybrid approach of Neural Machine Translation (NMT) and human-in-the-loop techniques to suggest domain specific translations of technical content to assist human translators. The current translation models are built using NFPA content datasets like Code & Standards, Applied Research, and Public Education and Outreach material and focus on bidirectional English-Spanish translations. The UI front end has options for you to get instant translations and edit incorrect ones suggested by the NMT machine.
  • LocationTools: Designed to run either on a single workstation or on a Cloudera cluster, this platform enables the fast, bidirectional geocoding of extremely large free-text datasets of street address information using publicly available information. LocationTools provides an easy to use API, multiple Docker-based deployment options, and thorough documentation of our innovative solution to the bulk geolocation problem using a combination of Natural Language Processing, efficient text indexing and searching, and Hadoop-enabled distributed computing.
  • Crowd Counter (OCC):A prototype platform that uses deep learning techniques to estimate and track crowd sizes from video imagery and compare the results to occupancy-specific limits derived from NFPA 101 Life Safety Code®.Initially funded by a NIST grant and developed in conjunction with the Fire Protection Research Foundation, OCC incorporates multiple neural network architectures and an easy to deploy web front end to visually monitor results in near-real time.
  • National Estimates Workflows: A library of transparent, documented, and replicable KNIME scripts used to produce NFPA’s National Estimates of the fire problem in the US by analyzing data from NFPA’s own annual Fire Loss survey and USFA’s National Fire Incident Reporting System (NFIRS).

NFPA researchers have also developed two web-hosted tools that take an innovative approach to addressing some specific fire safety issues.  While these applications are not open-source and are not under active development, we are providing free access to them and welcome any comments or suggestions for their future use.

  • Exterior Façade Fire Evaluation and Comparison Tool (EFFECT): A hosted tool intended for use by Authorities Having Jurisdiction (AHJ) to assess a portfolio of high-rise buildings where there is a concern that the exterior facade systems include combustible materials.  The tool aids AHJs in prioritizing buildings in their jurisdiction, conducting initial fire risk assessments of each building, and identifying those building that have a highest priority for inspection.  EFFECT is based on a Fire Risk Assessment methodology developed by Arup with peer review and technical input from Jensen Hughes.
  • Property Inspection Prioritization (PIP): PIP's aim is to assist the prioritization of property inspections by combining assessments of several risk factors with an underlying data science model derived from inputs from over 100 AHJs to "replicate" the prioritization of experienced inspector.  PIP, in other words, is an attempt at harnessing the collective wisdom of AHJs to produce a prioritization of properties that would roughly match what an actual AHJ would develop.