Our project explores the resilience of criminal networks, specifically for 5 different types of disruption strategies where a node is removed from the network and for 3 types of recovery strategies where the network tries to replace the removed node with another having the same role. Our work closely follow the paper "The Relative Ineffectiveness of Criminal Network Disruption" by Duijn P. et al.
The Jupyter notebook file makes use of functions imported from funcs.py and generates & simulates a criminal network which is disrupted at every time step specifying certain disruption and recovery strategies, and plots how the effieceincy, density, and average distance for new connection changes over time (all explained more thoroughly below). It should be noted that for the speed of the programm we only simulate one combination of disruption and recovery strategy at the time in these files, but for the graphs of the slides we did simulated them all together.
The slides from our presentation can also be found as a pdf here.
The initialize_network function makes use of the two .csv files in /Data that were taken from Table S-3 and Table S-4 from the paper's Appendix. It tries to rebuilt the cannabis cultivation value chain network to match the data as much as possible. From Table S-4 it takes the number of nodes (N) and average degree (D) per role and build a barabasi albert graph (n = N, m = D/N) for every role, then takes the disjoint union of these graphs and for every pair of roles (graphs) in this network, it adds the number of edges specified in Table S-3. Returns the value chain network, called G_VC.
add_macro_basic adds to G_VC a somewhat random outer network, using barabasi albert (n = 2000, m = 1). It loops over every node in G_VC and add some edges from it to the outer network. For the more connected roles Financing, Coordinato, and Growshop owner, it adds 1-10 edges, otherwise it adds 1-2 edges. Returns G_macro (only outer network) and G_combines (the new network with G_VC connected).
add_macro_stats adds another version of the outer network, this time based on the network statistics in Appendix (didn't use this in simulations as it didn't produce expected results). It multiples the data from the two tables by a factor of 3 and uses the same process as in initialize_network to build an outer network. G_VC and this outer network are connected in a similar way as in add_macro_basic.
network_stats prints and returns the number of nodes, number of edges, average degree, avgerage shortest path, diameter, and largest component of a given graph. If given a list of roles, it returns a dataframe with number of nodes, number of edges, and average degree for each role.
get_nodes_degree returns a dictionary of all nodes of a graph and their degree (to use in removing node with highest degree).
links_removed_node returns all the links of a removed node.
giant_component_perc calculates the percentages of the nodes that belong to the giant component.
calc_efficiency measure efficiency of the graph.
calc_efficiency_macro measures efficiency of the macro graph, calculating the distances of the nodes in G_VC through the macro network.
VC_degree_attribute adds VC degree as attribute to each node, calculated as the number of unique roles a node is connected to.
VC_degree_attribute_weighted similar as VC_degree_attribute but assigns weights according to how high in the hierarchy the role is (Figure 3 in paper).
removal_random removes a random node.
removal_highest_degree removes node of highest degree.
removal_highest_betweeness removes node with highest betweenness centrality.
disrupt_VC_degree removes node at random from set of nodes with highest VC degree.
disrupt_VC_degree_weighted removes node at random from set of nodes with highest weighted VC degree.
disrupt_VC_role removes node at random from set of nodes with specific role (in paper 'Diverting electricity').
random_recovery for every orphaned node, with probabilty of rewiring, finds possible replacements with removed node's roles, and chooses one of them randomly.
degree_recovery for every orphaned node, with probabilty of rewiring, finds possible replacements with removed node's roles, and chooses one with highest degree.
distance_recovery for every orphaned node, with probabilty of rewiring, finds possible replacements with removed node's roles, and chooses one with shortest distance.
simulate generates the G_VC graph then runs the specified disruption strategy n times, and at each time step with some probability of rewiring, uses the specified recovery algorithm. It calculates the percentage in giant component, efficiency and density at each time step.
simulate_macro_VC generates the G_VC and macro graph then runs the specified disruption strategy n times, and at each time step with some probability of rewiring, uses the specified recovery algorithm. It calculates the percentage in giant component, efficiency (macro version) and density at each time step.