Comparison of Decision Tree and Logistic Regression Models for Utilization in Sexual Assault Kit Processing

Area of Science:

• Forensic Science
• Genetics
• Criminology

Background:

• The significant backlog of untested sexual assault kits (SAKs) presents a challenge for forensic laboratories.
• Identifying factors that predict DNA profile quality is crucial for efficient processing and database inclusion.

Purpose of the Study:

• To identify variables impacting the likelihood of obtaining a probative DNA profile from SAKs for the Combined DNA Index System (CODIS).
• To develop and validate statistical models for predicting CODIS eligibility from SAK data.

Main Methods:

• Exploratory data analysis of Ohio's SAK Testing Initiative data.
• Validation study employing new and modified statistical models, including decision trees.
• Descriptive statistics to confirm variable relationships.

Main Results:

• Confirmed that the time between assault and kit collection, victim age, and recent consensual sexual activity are significant predictors of CODIS-eligible DNA profiles.
• The decision tree model demonstrated the highest predictive accuracy for CODIS eligibility.

Conclusions:

• Key variables reliably predict the success of DNA profiling for CODIS submission from SAKs.
• Statistical modeling, particularly decision trees, can optimize resource allocation in forensic DNA testing.