Title: EXAMPLE OF INTELLIGENCE MANAGEMENT SYSTEM THROUGH FORENSIC PROFILING: DRUG PROFILING

Example of intelligence management system through forensic profiling: drug profiling

Olivier Ribaux and Sylvain Ioset, University of Lausanne 

The systematic chemical and physical analysis of illicit drugs seized by law enforcement agencies has greatly developed since the middle of the nineties  . Illicit substances are seized, transferred to laboratories, and analysed in order to extract a “profile” (list of chemical substances and their quantities). The profiles are then recorded into a data base which is exploited in an intelligence or investigative perspective. For instance the process of linking illicit substance seized in different circumstances may lead to concentrate attention to a specific organised network while they were previously the object of separated investigations. Other indications about cultivation (origin), manufacture processes, or the distribution process of illicit drug trades can be inferred through the systematic analysis of the data base.

The data is organised into a dynamic memory: seizures are not stored individually but are rather collated and grouped into “classes” mainly according to similarity measurements between profiles coming from different seizures . Depending on which basis they are formed, these clusters mainly indicate similarities in the traffic at different levels, from the cultivation (origin) to the distribution of the illicit substance.  

Beyond standard clustering methods, other original methods for detecting patterns have been tested, particularly through spatio/temporal and graph visualisations. For instance, combinations of cutting agents are often used by drug smuggler before the distribution in the street. The spatio/temporal evolution of these co-occurrences inform on the dynamics of the local market . 

However, there is evidence that each drug trafficking network and laboratory develop its own receipts and methods that reflect differently into the intrinsic structure of the chemical profiles (correlations between variables). Thus, there is no suitable universal metric that can be defined, except for those specificities, and can systematically provide the same reliability when measuring proximity between samples. There is a need for a typical learning process as “classes” or specific groups profiles evolve over time, and show an inherent structure that may in turn influence the classification of new data.  

This hypothesis has been tested with data coming from known solved cases. Spectral clustering and its variants have been chosen to train the system and have shown to substantially improve the classification process . How those ideas may lead to the development of unsupervised methods is now the object of further developments.  

However, even if comprehensive European projects have led to some harmonisation and extension in the use of the method, in particular in the field of amphetamines , far from the whole potential of the approach being exploited. In fact, the central question is how to integrate knowledge extracted from drug profiling data bases with the analysis of other (traditional) sources of information (geopolitical, coming from investigations, etc.). Full aggregation of data, even theoretically ideal, can now be difficult to imagine as organisations that deal with the set of data are different (mostly forensic laboratories and the police), cover different countries and are based on different specialities. A more pragmatic model consists in the development of communication channels between partners organised as a network. For instance, chemical links can be systematically provided to the police and used in the investigative process. Conversely, investigative hypothesis can be tested through chemical profiling . This integration process must attract much more attention than the lack of communication between the organisations actually allows in practice (police, forensic laboratories and Universities).