Machine learning using random forest to differentiate between blow and fall situations of head trauma.

Abstract

Blunt head trauma is a common occurrence in forensic practice. Interpreting the origin of craniocerebral injuries can be a challenging process, particularly when it comes to distinguishing between falls or inflicted blows. The objective of this study was to develop a predictive model using an innovative Random Forest (RF) classification approach to differentiate injuries caused by falls from those caused by blows. The study examined 65 cases of blunt head trauma over the age of 18 resulting from a fall or an inflicted blow. A preliminary univariate logistic regression analysis followed by RF classification was performed. The presence of a depressed fracture and the lateralisation on the left-sided of cranial vault fractures, as well as extra-axial bleeding, in particular an extra-dural haematoma, were indicative of inflicted blows. The RF classification provided a simple predictive model with an accuracy rate of 78% to identify the most relevant injury criteria for distinguishing between falls and assault situations involving blows.