Characterizing Female Firearm Suicide Circumstances: A Natural Language Processing and Machine Learning Approach

Area of Science:

• Computational social science
• Public health informatics
• Forensic science

Background:

• Female firearm suicide rates have significantly increased since 2005, surpassing male rates.
• Understanding the circumstances preceding female firearm suicide is crucial for prevention efforts.

Purpose of the Study:

• To develop and evaluate a natural language processing (NLP) pipeline.
• To identify common circumstances preceding female firearm suicide using narrative data.

Main Methods:

• Manual coding of unstructured narratives from 1,462 cases in the National Violent Death Reporting System (2014-2018).
• Tuning of machine learning models (Naive Bayes, Random Forest, SVM, Gradient Boosting) using cross-validation.
• Performance assessment using metrics like specificity and positive predictive value.

Main Results:

• The NLP pipeline successfully identified key antecedents such as interpersonal disputes and intimate partner issues.
• Specific models demonstrated high performance, e.g., Gradient Boosting achieved 98.7% specificity for classifying interpersonal disputes.
• The pipeline showed strong positive predictive value in identifying suicide precursors.

Conclusions:

• A novel NLP pipeline was developed to classify female firearm suicide antecedents from narrative reports.
• This approach offers a potentially more efficient method for analyzing suicide circumstances compared to manual review.
• Findings can inform targeted prevention strategies and further research in suicide prevention.