Infrared spectroscopy combined with interpretable machine learning for forensic identification and weathering time estimation of empty blow fly puparia.

In this work, spectral data was used to identify the species and weathering time of blow fly puparia, which helps with the estimation of the minimum post-mortem interval (PMImin) in forensic science. After a preliminary model selection stage, the two best-performing modeling approaches were compared: a classical Logistic Regression (LR) model and a more complex Convolutional Neural Network (CNN). By evaluating both models on these two tasks, we aim to assess the performance trade-offs between competitive traditional machine learning approaches and modern deep learning techniques. Moreover, beyond comparative performance results, an interpretability analysis was conducted to investigate which regions of the spectrum were most informative for weathering time estimation and species classification. These insights may guide future research in forensic entomology using IR spectroscopy by highlighting spectral regions associated with key biological changes related to ageing and species differentiation.