Combining spectrum and machine learning algorithms to predict the weathering time of empty puparia of Sarcophaga peregrine (Diptera: Sarcophagidae)

Abstract

The weathering time of empty puparia could be important in predicting the minimum postmortem interval (PMImin). As corpse decomposition progresses to the skeletal stage, empty puparia often remain the sole evidence of fly activity at the scene. In this study, we used empty puparia of Sarcophaga peregrina (Diptera: Sarcophagidae) collected at ten different time points between January 2019 and February 2023 as our samples. Initially, we used the scanning electron microscope (SEM) to observe the surface of the empty puparia, but it was challenging to identify significant markers to estimate weathering time. We then utilized attenuated total internal reflectance Fourier transform infrared spectroscopy (ATR-FTIR) to detect the puparia spectrogram. Absorption peaks were observed at 1064 cm⁻¹, 1236 cm⁻¹, 1381 cm⁻¹, 1538 cm⁻¹, 1636 cm⁻¹, 2852 cm⁻¹, 2920 cm⁻¹. Three machine learning models were used to regress the spectral data after dimensionality reduction using principal component analysis (PCA). Among them, eXtreme Gradient Boosting regression (XGBR) showed the best performance in the wavenumber range of 1800–600 cm⁻¹, with a mean absolute error (MAE) of 1.20. This study highlights the value of refining these techniques for forensic applications involving entomological specimens and underscores the considerable potential of combining FTIR and machine learning in forensic practice.