A BDI Neuro-Symbolic Reasoner to Support Crime Investigation in Digital Forensics

Guilherme Dall'Agnol Deconto, Leonardo dos Santos Teixeira, Roben Castagna Lunardi, Rafael C. Cardoso, Felipe Meneguzzi, Avelino Francisco Zorzo

Keywords: Neuro-Symbolic Reasoner, BDI Agent, Digital Forensics, IoT

Abstract: The increasing presence of IoT devices has transformed buildings into smart environments, where sensors continuously record contextual information. Incomplete or inconsistent data can compromise the reconstruction of events and prevent the production of reliable evidence, creating significant challenges for criminal investigators working in digital forensics. Pure machine learning solutions struggle with the problem of generalisation beyond the training data. This motivates the exploration of neuro-symbolic approaches, which combine the efficiency of machine learning models with the adaptability of symbolic reasoning. This paper develops a hybrid architecture where machine learning components extract low-level events (e.g., room occupancy, sensor anomalies), and a BDI agent reasons over them to identify criminal patterns in smart environments. We empirically evaluate an implementation of the resulting architecture in a forensics scenario under increasingly degraded sensor conditions, showcasing its robustness.

Area: Engineering and Analysis of Multiagent Systems (EMAS)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 285