Improving the usability of large emergency 911 data reporting systems: A machine learning case study using emergency incident descriptions

Abstract

Introduction: Emergency 9-1-1 incident data are recorded voluntarily within fire-department-specific computer-aided dispatch systems. The National Fire Incident Reporting System serves as a repository for these data, but inconsistency and variability in reporting practices across departments often lead to challenges in data quality and utility. This study aims to enhance emergency incident categorization and explore the feasibility of an automated system using free-text incident data from the National Fire Operations Reporting System (NFORS). Method: Researchers extracted and standardized 3,564 unique 9–1-1 incident descriptions from six fire departments using NFORS data, including narrative fields from emergency reports. The data were preprocessed using natural language processing (NLP) techniques, such as tokenization, stop word removal, and feature extraction (e.g., TF-IDF and n-grams). These features were used to train and evaluate Machine Learning (ML) models, including Naïve Bayes, Random Forest, and Support Vector Machine, to classify incidents into nine categories. The NLP techniques prepared the text data for the ML models, which performed the classification and assessed the automated system’s performance. Results: The study demonstrated significant improvements in incident categorization accuracy using the NLP and ML approach. Unigram models achieved 93% accuracy when applied to 3,564 unique incident descriptions. This performance was evaluated by comparing the automated classifications to manually assigned categories, which served as the reference. Mis-categorizations primarily occurred with “Emergency Medical Services (EMS).” Conclusions: Standardized and consistent incident categorization is vital for informed decision-making, efficient resource allocation, and effective emergency response. Our findings suggest that adopting a robust categorization system, such as the nine-category model using NLP and ML, can improve categorization accuracy and enhance data quality and utility for decision-making. Practical Applications: Public safety agencies can leverage these insights to modernize data systems, strengthen occupational surveillance, and create more resilient and sustainable public safety data systems.