Hi All, I am Nadeem Anjum, a third year bachelor's student of the Department of Computer Science and Engineering, Indian Institute of Technology (IIT) Kharagpur. I am well versed in Java, XML, PHP, MySQL, JavaScript, jQuery, HTML, CSS, C, C++ and Python. I have been an active contributor to numerous development and open-source projects: http://cse.iitkgp.ac.in/~nanjum/OpenSourceProjects.html . I have also been involved in a number of research projects: http://cse.iitkgp.ac.in/~nanjum/#Research
I have been in touch Prof. Suresh Marru regarding the GSOC project: https://issues.apache.org/jira/browse/SIS-97 I have done the required background reading for the proposal and want to share my ideas for your feedback. __________________________________________________________________________________ *Embedding Google Maps and randomizing agent moves:* We shall use google maps API and the location of an agent will be denoted by the latitude-longitude. The agent will be able to move along the roads on the maps only. The houses/buildings will be the sites the criminal agent (c-agent) can break in. Each house/building will be associated with a victim-agent (v-agent). *Randomizing agent moves*: The c-agent will have a predefined starting position (lat-long). Along a single straight road, the agent can choose to move in either direction with a random probability of 0.5 each. Across a road junction having x roads, the agent can choose to move along any road with a random probability of 1/x. *Simulating Distance Decay Theory:* The agent starts with an initial equity and is deducted a mark for every y meters (say 100 meters) the agent travels. This ensures that the agent does not go too far away from its home location. Each house has a profit value associated with it and a level of acquaintance of the v-agent with a c-agent. On coming across each house, the agent decides with a random probability based on the above two factors, whether or not to break into the house. On a successful break-in, the agent's equity will be increases by the profit amount of the house and the agent returns to his home. The agent is designed to return to his home following the shortest distance from any location. The program will be designed to ensure that the agent will decide to go back, so that it is able to reach home without spending all its money. To ensure that the agent does not go on a crime spree, we make the agent turn back immediately after breaking in a cell and put a maximum amount that can be spend in travelling (say around 10% of the initial equity). *Memory Function: *The c-agent will refrain from committing crime in its immediate neighborhood, i.e. with the v-agents it is acquainted with. We shall define a circular area of say z meters square (say 100 meters square) and the c-agent shall recognize all v-agents falling into this circle and vice-versa. However, we also need to take into account that the agent forgets those who he has not met for a long time. So if the v-agent passes through a c-agent at time T0, at time Tn, the probability that the c-agent will recognize the v-agent and vice-versa will be 1-f(n), where f(n) increases wrt n. Say f(n) = 0.01*n, where n is the distance travelled in meters (assuming agent moves at constant speed of 1 m/s). This means that after travelling say 50 m, the probability that the c-agent recognizes the v-agent is 50%, and the memory will be active upto 100 m . We also need to take into account the fact that if the agent sees a person repeatedly, it will remain in memory for a longer time. For this, we modify f(n) as f(x,n)=x*n. Initially x=0.01. This means memory is active for 100m. If within 100m, the c-agent again interacts with the v-agent, we change x to 0.9*x. This will ensure that the criminal will be remembered more than 100m (long-term memory) The agent will break into the house with a probability which is: 1. directly proportional to the profit value of the house 2. indirectly proportional to the probability that the v-agent recognizes the c-agent. If the resident v-agent has the c-agent in its long-term memory, the c-agent is caught, its equity drops below a certain threshold and the program restarts again. We shall run this simulation a large number of times. *Simulating Routine Activity Theory*: We can associate with each house/cell a fraction p, which denotes the ease of breaking into the house. For example for a well-secured house with security cameras, lights, fences, strong locks etc., the ease of breaking in will be very low. On the other hand for an open park, the ease of breaking in will be high. The c-agent will now choose to break into a cell with a probability which is: 1. directly proportional to the profit value of the house 2. indirectly proportional to the probability that the v-agent recognizes the c-agent. 3. *directly proportional to the ease of breaking into the cell, p* * _______________________________________________________________________________________________ * Please provide your suggestions and feedback before i formalize my proposal. Thanks, Nadeem Anjum
