2. What is Ship Happens?
Our model predicts, yes/no, whether there is an elevated level of risk for injury onboard a
commercial ship compared to historical ship injury data; having such information can help inform
asset allocation and enforcement efforts while reducing risk to personnel
Our data sources include AIS (Automated I.D. System) data and MISLE (Marine Information for Safety
and Law Enforcement) data—both data sets were provided by the US Government (US Coast Guard)
for research purposes
3. If we determine a correlation between ship characteristics and injury events using historical
data and a model then we can predict incident risk.
Successfully doing so may inform asset allocation and enforcement efforts while reducing
risk to personnel.
Hypothesis
6. Data Source
Marine Casualty and Pollution Database from Marine Information for Safety and Law Enforcement (MISLE) site
(https://www.marad.dot.gov/resources/data-statistics/)
WORM Storage
Postgress Database
Tables Used
− MisleVessel
− MisleActivity
− MisleInjury
Attributes Used
(Initially) activity_id, gk_d_vessel, vessel_id, gross_ton, vlength, vdepth, vessel_class, vessel_age, flag_abbr,
route_type, mvaccident, incident_yr, incident_mo, incident_dy, accident_type, casualty_type_desc
Ship Happens: Data Ingestion
7. Left outer join
− MisleVessel
− MisleActivity
− MisleInjury)
Removed duplicate data
Initial set
− Instances: 1353830
− Features: 16
Final set
− Instances: 260474
− Features: 7
− gross_ton; vlength; vdepth; vessel_class; vessel_age; route_type; mvaccident
Ship Happens: Data Wrangling
10. Features We Selected
− Gross ton: size/type of ship
− Vessel Length: restricts routes able to be taken and ability to travel
− Vessel Draft: restricts routes able to be taken and ability to travel
− Vessel Type (Classification) : type of ship has various risks according to usage
− Vessel Age: age of ship changes features/safety/durability/reliability
− Route Type: different travelled routes have different risks
− Accident Detail: type of accident, number of injuries/deaths, place and time of accident
21. Application Scenario
US maritime missions (i.e., police departments,
US Coast Guard, Fish and Wildlife Services):
− Safety of navigation
− Search and rescue
− Anti-drug/anti-human trafficking
− Anti-poaching/fisheries regulations
− Anti-pollution enforcement and monitoring
− Criminal investigations
Many potential violators, few resources –
reducing budgets, increasing requirements
Scenario: Port of Baltimore, common
operational picture (COP), 14 January 2017
− Data available: live streaming, 24-hour
period, etc
− Ship positional data – 19 commercial ships
22. Scenario Outcome
Out of 19 ships present in the Port of
Baltimore on 14 January, 0 were
determined high risk by our model.
− Possibly a low risk port;
− Possibly only low risk vessel types;
− Perhaps indicates underreporting;
− Model weakness
With additional improvements to the
process and model, operators at a port may
use this information as a type of tipping
and queuing for operational planning day-
to-day, or year-to-year, for future planning
23. Way Forward: Challenges
− Improvements to model
− Misinterpretation of outputs: too much/too little confidence in the tool, inability to use
effectively
− Data quality: bad (inaccurate or falsified) data or unreported data will impact confidence
in accuracy of outputs
− Data flow/ingestion: new, validated data to be able to adjust to changing factors (ex:
increases in crime at a particular port, decreases in risk due to canal expansion, etc)
24. Way Forward: Changes
Features
− Time: certain times can be more at-risk than others (ex: 4th of July)
− Location: environmental risks (ex: narrow waterways vs draft restricted waterways,
underwater features)
− Weather: shipping safety is weather dependent
− Operators: individuals or organizations more/less risk
Data
− Volume of data: incident specific
− Variety of data: sources and types (weather, owner, operator, charter, bridge crew, deck
crew, geographic data, time/date data, navigational aids, hazards to navigation