Aim of the project
To support investigations by conducting behavioral analysis, analyzing motion patterns and call data records of individuals under inspection. Via utilizing cell tower based information and calling data record analysis, investigators will have the ability to uncover potential collaboration between pairs of individuals for a given suspect event or expand/reduce scope of investigation to individuals likely to have been on suspect scene.
Methodology
Behaviors, motion patterns, suspect event related communication will be uncovered from location based analysis of individuals' phones being registered on neighboring cell towers, and this information will be combined with call data record based calling circle and community analysis. Initially, the period under analysis will be divided into smaller time units, in which the central indicators will be concluded as if they were static for the given units of time periods. This will help investigators to get a high level picture of the suspect related information. During this static analysis of time intervals, categorical locations will be defined that will help investigation:
- Crime/event related objects - e.g. crime scene, escape path, etc.
- Investigation related objects - e.g. living address (cell tower), favorite pub/restaurant, place of work, friend/acquinantance place of living, parking space, etc.
- Public objects - e.g. restaurant, post, bank, pharmacy, police station, cinema, shopping mall, etc.
After these central points are determined for relevant objects, a circle of radius will be defined per person, from which likelihood of space overlap between suspected collaborating pairs of criminals can be calculated.
The assumed precrime suspect pathway will be revealed. Probabilities for physical crosses of precrime suspect pathways will be assigned. These location based information are combined with Call Data Records of suspected pairs of individuals whose motion patterns are similar even though their base center and typical places of their pathways are differring.
Via applying calling circle and community analysis, investigation suspect list can be expanded/reduced by applying learning algorithms. Investigation officers can interactively support whether a given crime fact can be excluded or not. The manually validated or rejected scenarios will also affect probabilties assigned to the remaining movement patterns through. The learning algorithm will rank the potential scenarios based on probability, and further manual validation/exclusion of likely scenarios can be done by considering additional evidences and information.
Results
The main deliverables of projects are:
- Probability of being present on scene at time of event
- Probability of preliminary analysis of the (crime) scene before
- Meeting probability between suspected pairs of collaborators
- Most probable meeting object (apartment, pub, parking, etc.)
- Probability of moving together on same route (at same time or with time distance)
- Calling pattern analysis of suspected collaborators
- Community identification of suspected collaborators up til 3 degrees of separation to expand suspect lists
- Likelihood of other calling card IDs or mobile subscriptions belonging to same person
DOs and DON’Ts
Do
- Combine cell tower information with Social Network Analysis
- Use fingerprinting algorithm on call data to see if suspect uses multiple calling cards reaching out to same relationship network
|
|
Don’t
- Assume that the AI adjusted learning algorithm alone can assume solutions without many other science and logical inputs into the case
|
The figure shows graphical aids that can help investigation by representing weighted links between individuals whose mobile were logged on nearby cell tower information in the past couple of months, revealed in a time series motion pattern analysis. With the help of adding coordinates of categorical locations that are relevant to investigated events, potential lists of pairs of suspects can be better identified. This can be done for example concluding whose motion patterns follow similar characteristics around scene even though their base center and residency/workplace might be outside of crime scene) A meeting probability (in time and space) can be also calculated for each pairs.
Why we are different
- Experience with real public intelligence projects
- Delivered application is used for crime analysis
- Social Network Analysis expertise of mining and analyzing complex graphs gives a clear capability differentiation
|
