IsoProb - geographical assignment systems of criminal traces based on spatial machine learning and isotopic modeling

• Grant number: 22/07086-7
• Support Opportunities: Regular Research Grants
• Start date: November 01, 2022
• End date: October 31, 2024
• Field of knowledge: Interdisciplinary Subjects
• Agreement: FACEPE
• Principal Investigator: Luiz Antonio Martinelli
• Grantee: Luiz Antonio Martinelli
• Host Institution: Centro de Energia Nuclear na Agricultura (CENA). Universidade de São Paulo (USP). Piracicaba , SP, Brazil

Abstract

It is expected that an efficient isotopic geographic origin attribution model (MIAR) is one that presents a structure based on (I) a geographically comprehensive and structured database, (II) that has as input base information that encompass local, global and spatially dependent variations, (III) that present isoscapes built both on more accurate "machine learning" (ML) techniques and that incorporate the spatial autocorrelation of the data; and (IV) that allows, in an adequate and efficient way, the modeling of the geographic attribution from one or more isotopes simultaneously in a generalized way and that incorporates possible correlations between the marginal distributions. In this way, the present project has as main objective the construction of an efficient system of geographic allocation of samples of criminal traces for the Brazilian territory based on these premises, and that can be effectively used by national and state security agencies. For this, the project will rely on the long experience of the Center for Nuclear Energy in Agriculture at the University of São Paulo (CENA/USP), more precisely, with the team of Prof. Dr. Martinelli (Isotopic Ecology Laboratory) in the collection and analysis of samples to generate isotopic data, in collaboration with the group of researchers who are part of the faculty of DEINFO and the Program in Biometrics and Applied Statistics (PPGBEA) at UFRPE. DEINFO members have great expertise in solving complex problems of nature and society, from the development and application of statistical techniques, ML, Geostatistics, and computationally intensive models. In addition to this partnership, the project will have the participation of members external to FACEPE/FAPESP and who have a direct interest in the elaboration of the IsoProb, such as Prof. Dr Gabriela Nardoto (President of RENIF), from the criminal expert of the Federal Police Dr. Fábio Costa, and Dr. João da Silva Sena, who published one of the first works with the application of ML for the construction of Isoscapes. To achieve the general objective, the project will be divided into four stages: I - Structuring an isotopic database; II - Construction of matrices of integrated general and spatial "features"; III- Machine Learning in the development of Isoscapes for the Brazilian territory, and IV - Mathematical models of geographic origin attributions. The IsoProb system can be used both from a scientific point of view and as a tool by national and state security agencies for forensic investigations. It is expected that, from georeferenced samples obtained in Federal Police field operations, for example, their isotopic values can be compared with the predicted values in the Isoscapes contained in the IsoProb, and thus generate a continuous map or nominal blocks of probability of attribution of the origin of those samples. (AU)