Explainable deep learning in bloodstain pattern analysis: A pilot study using convolutional neural networks with...

Area of Science:

• Forensic Science
• Artificial Intelligence
• Machine Learning

Background:

• Bloodstain Pattern Analysis (BPA) is crucial in forensic investigations.
• Current manual classification methods can be time-consuming and subjective.
• Advancements in artificial intelligence offer potential for objective and efficient analysis.

Purpose of the Study:

• To assess the feasibility of applying a novel explainable deep learning (XDL) methodology to classify Bloodstain Pattern Analysis (BPA) patterns.
• To investigate the use of convolutional neural networks (CNNs) for classifying impact and non-impact BPA patterns.
• To introduce saliency maps as an explainability layer for XDL in BPA.

Main Methods:

• A convolutional neural network (CNN) was trained using a combination of researcher-generated and open-source BPA datasets.
• The CNN was employed to classify impact and non-impact BPA patterns.
• Saliency maps were utilized as the explainability layer for the CNN model.

Main Results:

• The XDL methodology achieved up to 79% accuracy over 10 folds in classifying BPA patterns.
• The application of saliency maps demonstrated a novel use of XDL in BPA interpretation.
• The results validate the feasibility of using AI for BPA classification.

Conclusions:

• Explainable deep learning presents a promising new research direction in BPA.
• The developed model shows potential as a reliable tool to assist BPA experts, potentially expediting interpretations.
• This approach can enhance transparency and trust in AI-driven forensic analysis, improving reliability.