Converted for use in DELVE by Radford Neal, June 1996.
Originally compiled by Robert Dawson, 1995.
The titanic dataset gives the values of four categorical attributes for each of the 2201 people on board the Titanic when it struck an iceberg and sank. The attributes are social class (first class, second class, third class, crewmember), age (adult or child), sex, and whether or not the person survived.
The question of interest for this natural dataset is how survival relates to the other attributes. There is obviously no practical need to predict survival, so the real interest is in interpretation, but success at prediction would appear to be closely related to the discovery of interesting features of the relationship. Note that there are only sixteen possible combinations of input attributes for this prediction task, so the interesting behaviour will be that with small training sets.
Carriage returns at the end of the lines were deleted, as was a line containing a period at the end of each file. Other than this, the titanic.doc and titanic.dat files are as obtained from this source.
The dataset was compiled by Robert J. MacG. Dawson, and discussed by him in the on-line article 'The "Unusual Episode" Data Revisited', Journal of Statistics Education, vol. 3, no. 3 (1995), available via the URL above.
As discussed in the article, the dataset was reconstructed from sources that were not completely clear, so there are undoubtably some errors.
The cases in titanic.dat are clearly in a non-informative order, grouped by identical attribute patterns. This has been retained for the DELVE dataset file.
The representation of attributes has been changed to be more mnemonic.
Prior information regarding the significance of social class is somewhat debatable. In the standard prior, I have considered status to be an ordinal variable in which crewmembers come after third class passengers. Perhaps crewmembers should be considered to be outside this class ordering altogether, but that is not convenient.
Last Updated [Date]
Comments and questions to: email@example.com