Sea Turtle Rescue: Error Detection Challenge

Help Local Ocean Conservation clean their sea turtle rescues database
About Local Ocean’s sea turtle rescue program
Since 1998, Local Ocean Conservation (LOC) has been running a by catch net release program. By catch is when a non-target species, in this case endangered and critically endangered species of marine turtles, are captured in fishing gear accidently. They become tangled in nets and risk injury, drowning and even slaughter. What do we do about it? We work with around 350 local fishermen who, instead of slaughtering the turtles they catch or leaving them to die, contact us. We are then able to race to the rescue!
We assess the turtles condition, checking for parasites and injuries.
We collect data such as measurements, weight, species, gender etc.
We attach a tag or record the number of an existing tag
We maintain a database of all recorded data
We monitor the condition of turtles that have been recaptured, sometimes these are even ex-patients from our Rehabilitation Centre!
If the turtle is fit and healthy we transport them to a safe release site where they are returned to the ocean. If the turtle is sick or injured, we transport them back to our Rehabilitation Centre for medical treatment and provide specialist care until such time as they are strong enough to be released.
We provide a small amount of remuneration to help participating fishermen to repair any damage to their gear and cover any other expenses incurred such as telephone and transport costs.
Each year the number of fishermen involved in this programme increases along with the number of turtles released, a reflection of the success of the linked education and community development programmes.
Since 1998 over 10,000 turtles have been released through this programme.
The problem YOU will help us solve!
Each time a fisher catches a turtle he/she delivers the turtle to LOC researchers who collect the data on the rescue in a handwritten logbook. (Though we hope to introduce an electronic form of data collection soon.) Back in the office, LOC staff and volunteers then enter the handwritten data in an Access database. At the end of the year, LOC staff review each rescue/field to compare the database entry with the hand-written log book to ensure the integrity of the data.
LOC would like to reduce the amount of time and effort needed for this review process. (And it is a lot!)
The objective of this challenge is to process the data in the turtle rescue database and assign a probability that any given field has been entered erroneously from the logbook into the database.
About the data
You are asked to make predictions for each field in the test dataset about the likelihood of it containing an error (i.e. that the cleaned data is different from the pre-cleaned data). You will train your model on all of the data from 1998 to 2010 (except for 2008 and 2009). The test dataset is the 2011 data.
dirty_data.csv is all of the data from 1998 to 2010 (except for 2008 and 2009) as they were originally entered into the database from the handwritten logbooks.
cleaned_data.csv is all of the data from 1998 to 2010 (except for 2008 and 2009) after they were reviewed and corrected as needed.
test_data.csv is the pre-cleaned (dirty) data from 2011.
submission example.csv is an example of what your submission file should look like. Note that the variable ID in the submission file is "Rescue_ID" + " x " + each column name. The order of the rows does not matter, but the names of the IDs must be correct. The column "error" is your prediction of the likelihood of the field containing an error. (0=No error, 1=Error)
variable_definitions.csv is the full list of variables and their explanations.
Note if the entry for a particular field in both the clean and dirty data is "NA", this should be assessed as no error.