Blows or Falls? Distinction by Random Forest Classification.

Area of Science:

• Forensic Anthropology
• Biomechanics
• Machine Learning in Medicine

Background:

• Distinguishing between accidental falls and intentional blows causing fractures is crucial in forensic and archaeological contexts.
• Current methods may lack the precision needed for accurate classification, especially in complex cases.

Purpose of the Study:

• To develop and evaluate a machine learning-based classification method for differentiating fracture origins (falls vs. blows).
• To assess the accuracy of random forest models using skeletal and demographic data.

Main Methods:

• Utilized a dataset of 400 anonymized patients (ages 20-49) with fractures from falls or blows.
• Employed random forest classification, testing various model parameters and feature encodings (anatomical regions, bones, age, sex).
• Optimized models based on binary coding of 12 anatomical regions or 28 bones, with and without baseline data.

Main Results:

• The best random forest model achieved an 83% accuracy rate in distinguishing between fractures caused by falls and blows.
• Model performance was sensitive to the selection of random forest parameters and feature encoding strategies.
• Binary coding of anatomical regions or bones, with or without age and sex, yielded the highest accuracies.

Conclusions:

• The proposed random forest classification method demonstrates significant potential for accurately differentiating fracture causes.
• This approach can serve as a valuable tool for forensic experts and archaeologists in interpreting skeletal trauma.
• Further refinement of machine learning models could enhance the reliability of skeletal trauma analysis.