A Deep Dive into TENSOR’s Technological Innovations: Behavioural Biometrics

Author: Quadible

Introduction

In the complex landscape of digital investigations, a smartphone found at a crime scene -an “orphan device”- is often a critical piece of evidence. However, linking that device to a specific individual (device owner) is frequently a significant challenge for Law Enforcement Agencies (LEAs). As part of the TENSOR project, an innovative solution to support this problem has been developed based on AI-based Behavioural Biometrics. This technology moves beyond what a user does on their phone to analyze how they do it, creating a unique digital “profile” that can help identify suspects.

Outline of Behavioural Biometrics technology

Behavioural biometrics is based on the principle that every individual interacts with their device in a distinct, measurable, and repeatable way. This “behavioural signature” is comprised of subtle, unconscious patterns. The TENSOR solution analyzes a wide range of these traits, including keystroke dynamics (the specific rhythm, timing, and pressure used when typing), stroke patterns (the velocity, pressure, and geometry of swipes and taps), and even the device angle (the way a person typically grips or orients their phone, measured by integrated sensors like the accelerometer). By also incorporating location points of interest from GPS or WiFi traces and general device usage patterns, the system fuses these data streams to build a unique behavioural profile.

AI algorithms are trained to learn a user’s unique patterns from each individual behavioural trait, and then the system fuses the results to calculate a final confidence score. A key innovation in our approach is how the system utilizes a one-shot learning approach; it can add new suspect profiles to the database individually without needing to retrain the entire model from scratch. This process makes the system highly scalable and efficient for LEA operations.

Added value to LEAs and TENSOR project

The primary value of this component lies in its ability to make a link between an anonymous device and a known individual. In investigations, LEAs may sometimes hit a dead end with “orphan” devices, and this behavioural biometric solution offers a new, supporting form of digital evidence. Within the TENSOR project, this component provides an additional layer of identity analysis that complements other forensic tools and biometric components, to assist LEAs towards a comprehensive understanding of the digital evidence they recover for suspect identification.

In practice, the behavioural biometrics is envisioned to work in two phases: enrolment and identification. During the enrolment phase, when a criminal is first arrested and processed, LEAs could (following legal protocols) have the individual perform specific, quick actions on a device running a specialized app. This captures their unique behavioural profile, which is then securely stored in a criminal biometric database, similar to how fingerprints or facial biometrics are stored today.

The identification phase begins when an orphan device is found and recovered. To derive the user’s behavioural profile from this device, through proper legal channels, LEAs could request the raw, anonymized behavioural data from the vendor of an installed app that incorporates behavioural data collection for internal security or user experience purposes (e.g. banking, gambling, health-related apps). This data, often in a simple structured format (e.g. JSON), can then be securely uploaded to the TENSOR platform.

Once uploaded, the Behavioural Biometric Identification component processes the raw data from the orphan device, generates its corresponding behavioural profile, and then compares it against the entire database of known suspect profiles. The system then returns a ranked list of potential suspects with detailed similarity scores.

Conclusion

In the TENSOR project, we are transforming advanced behavioural biometrics from a concept into a practical, supporting tool for law enforcement. This component enhances the investigative capabilities of police authorities, providing an AI-driven method to link suspects to “orphan” devices found at crime scenes. The vision is to help LEAs analyze evidence more effectively and set a new standard for how innovative behavioural biometric technologies can be responsibly applied to ensure public safety.