Skip to content
Risk model to improve construction related building safety
Jupyter Notebook
Branch: master
Clone or download
Latest commit efcb96e Aug 27, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Bad_Actor_Germain_3_Automation.ipynb Add files via upload Jul 19, 2018
README.md Update README.md Aug 27, 2019

README.md

Social Network analysis as a Risk Tool for Building Safety

The idea behind this project is that previous bad behavior is a good predictor of future bad behavior, and that network ties to bad actors will be a useful tool to categorize high risk construction permits for audit or inspection.

The Department of Buildings issues approximately 300,000 building permits a year, a number of these permits will result in illegal work. Each construction permit contains the names of the actors working on the project. These include the owner, architect or engineer, contractor and the filing representative. Each permit produces a network of actors working at a building in New York City.

A bad actor list is generated based on previous illegal work or dispositions.

This project had two steps. 1) Among the list of known bad actors, creating a "bad actor" ranking system. 2) Rank the most recent permits as high or low risk based on the "badness" rank of the actors working on the project (see below measures)

Risk Tool Measures:

Transactional

Total number of ties: Total number of ties each actor has in the network
Total number of bad ties: Total number of bad ties to each actor (are they working on a project with a known bad actor?)
Bad tie density = Total number of bad ties/Total number of ties

Bad Actor Work Relationships

Total number of unique ties: Total number of ties each actor has in the network
Total number of unique bad ties: Total number of bad ties to each actor
Bad unique tie density = Total number of unique bad ties/Total number of unique ties

Influence and power in the industry

Betweeness Centrality: Measure of actors who are powerful in the industry by brokering relationships between groups
Total Number of BINs: Total number of buildings they worked on
Unique Number of BINs: Unique number of buildings they worked on, each BIN is counted only once

Membership in “bad” community (community detection analysis)

Community Membership: The name of the community in which they belong
Community Bad Density Unique = # Bad Unique in Community/# Number Unique in Community

Badness Risk Calculation

Three of the above measures were used to rank the level of "badness" among the bad actors (about 500 bad actors in all): previous DOB violations, total unique bad tie density and betweeness centrality. Total uniqe bad tie density is a measure of how many ties they have in their network that are bad, i.e. on average the people they do business with tend to engage in illegal activity. Betwneeness centrality is a measure of the actor as a key person in the network who is conducting bad behavior among several groups. The actor is therfore central in dissemnating bad behavior in the industry.

Network analysis example (betweeness centrality)

Graph of the top 20 bad actors. The most central actor in the network (Shmiel Firth) recently received violation for building collapse. Yellow nodes are bad actors, white nodes are good actors (or unknown if they are bad).

badactor20network

Example of community detection, clusturing of bad actor communities:

badactor20communities

You can’t perform that action at this time.