sis-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jesse Maes (JIRA)" <>
Subject [jira] [Commented] (SIS-324) Develop Hidden Markov Model to predict criminal moments
Date Thu, 24 Mar 2016 19:04:25 GMT


Jesse Maes commented on SIS-324:

Hi Dr. Marru,

Sorry, I didn't realize the file would be readable before I actually submitted my comment!
I've converted the proposal to a Google Doc for the sake of sharing on the Summer of Code

I also went through and changed the pronouns, as you suggested. I consider documentation and
testing to be a part of development, and I language to try to make that more obvious.

> Develop Hidden Markov Model to predict criminal moments  
> ---------------------------------------------------------
>                 Key: SIS-324
>                 URL:
>             Project: Spatial Information Systems
>          Issue Type: New Feature
>            Reporter: Suresh Marru
> The proposed project extends
> The Agent based modelling project during GSoC 2013 used a probabilistic model that was
hardcoded. It served as the basis to predict criminal’s movements which infer crimes. This
proposed project should do the inverse. Predict the probabilistic model that controls the
criminals’ behavior using data about his movements and crimes.
> The project should be preliminary work for data mining. From a sample anonymised emergency
call (911 data) a criminal should be uniquely identified. Hidden Markov Model, a probabilistic
state transition system, i.e., we define states such as “at home”, “in office etc”,
“roaming mode” etc, and there are probabilistic transitions between them. We can associate
some behavior to a particular state. Thus, the probabilistic model that was hardcoded by Nadeem
(in GSOC 2013) was a Markov model (this is a little indirectly). When we only have the crime
data and we want to find the model that dictates the criminal’s behavior, the Markov model
is hidden to us.  There are algorithms that can do this. These algorithms need to have their
parameters set by humans – such as the number of states. So this would need some amount
of experimentation.

This message was sent by Atlassian JIRA

View raw message