okcupid_stem
OpenML dataset with id 45067
No author found.
Full work available at URL: https://api.openml.org/data/v1/download/22112025/okcupid_stem.arff
Upload date: 27 January 2023
Dataset Characteristics
Number of classes: 3
Number of features: 14 (numeric: 2, symbolic: 11 and in total binary: 1 )
Number of instances: 26,677
Number of instances with missing values: 0
Number of missing values: 0
User profile data for San Francisco OkCupid users published in [Kim, A. Y., & Escobedo-Land, A. (2015). OKCupid data for introductory statistics and data science courses. Journal of Statistics Education, 23(2).]. The curated dataset was downloaded from [1]. The original dataset was created with the use of a python script that pulled the data from public profiles on www.okcupid.com on 06/30/2012. It includes people (n = 59946) within a 25 mile radius of San Francisco, who were online in the last year (06/30/2011), with at least one profile picture. Permission to use this data was obtained by the author of the original paper from OkCupid president and co-founder Christian Rudder under the condition that the dataset remains public. As target, the variable 'job' was collapsed into three categories: 'stem', 'non_stem', and 'student'. STEM jobs were defined as 'job' %in% c('computer / hardware / software', 'science / tech / engineering'). Observations with 'job' %in% c('unemployed', 'retired', 'rather not say') or missing values in 'job' were removed. The factor labels of the variable 'speaks' had to be changed to integers to prevent a bug which would not allow the upload of the variable as a nominal feature. The original dataset also included ten open text variables 'essay0' to 'essay9', which were removed from the dataset uploaded here. The dataset further includes the date/time variable 'last_online' (ignored by default) which could be used to construct additional features. Using OkCupid data for predicting STEM jobs was inspired by Max Kuhns book 'Feature Engineering and Selection: A Practical Approach for Predictive Models' [2].
