Speaker: Manuel Fernando Mollón Laorca
Abstract: The growing adoption of AI in forensic science demands high performance, interpretability, robustness, and transparency. This research, part of the Horizon Europe Natural Traces Project (https://naturaltraces.com) advances responsible AI through two key forensic applications. First, infrared (IR) spectral data was used to identify blow fly puparia and estimate their age, aiding minimum post-mortem interval estimation, following up the results in [1]. Researchers compared classical Logistic Regression with Convolutional Neural Networks to evaluate performance trade-offs. Beyond accuracy, interpretability analyses identified the specific spectral regions driving model decisions, ensuring results align with biological reality. Second, transformer-based DNA embeddings were employed to classify forensic soil samples in the setup presented in [2]. To ensure robustness, various preprocessing strategies and model configurations were evaluated, with temporal inference models achieving strong results. These case studies illustrate how responsible AI frameworks strengthen forensic evidence by balancing predictive success with explainability, reliability, and scientific accountability.