Intelligent Bloodstain Pattern Analysis Using Convolutional Neural Networks

Bloodstain pattern analysis is a fundamental part of forensic crime scene reconstruction.However, the samples collected from real crime scenes are often obtained under inconsistent conditions, on different surfaces, and at various impact angles.This makes it difficult to classify them accurately.While research in this area has progressed from rule-based and manual interpretation methods to deep learning-based and convolutional neural network (CNN)-based models, there remains a significant lack of consistency in how their performance is evaluated.This paper presents a systematic review of 76 research papers, organising the field of bloodstain pattern analysis using CNNs into four methodological paradigms, and proposes a structured evaluation framework to address this evaluative gap.It is evident from this review that modellevel accuracy metrics dominate the literature, while real-world forensic validation remains rare, and controlled laboratory datasets are overrepresented, limiting their applicability in authentic casework settings.Moreover, this review highlights the need for standardised datasets reflecting real-world forensic variability and emphasises the integration of explainable artificial intelligence to improve interpretability and support legal reliability, thereby strengthening the practical applicability of CNN-based bloodstain pattern analysis.This paper argues that, unlike previous research, the priority for future work in this field must shift from achieving high classification accuracy under ideal conditions to establishing forensically grounded, statistically transparent, and forensically meaningful evaluation standards.