C-OBGET (OBGET for Children)
• Psychological dataset for children can be used to train machine learning models which can classify and detect mental disorders in childhood, and prevents the progression of the disorders. • We collected Original Bender Gestalt Drawing Test dataset for children for automating Original Bender Gestalt Drawing Test for children. • There is no psychological drawing test dataset for children with related metadata. To the best of our knowledge most researchers focus on computerized psychological tests for adults, while many mental disorders in adults, initiate childhood problems. Therefore, by detecting mental disorders in childhood, it is possible to prevent many irreparable mental disorders in adulthood. • Collecting children hand sketch dataset is very challenging, children spatial visualization is incomplete and they cannot draw shapes correctly. • Collecting, analyzing, and labeling children's hand sketches are harder than adult hand sketches because it takes more work to explain test rules to children. Children do not sit for a certain period and draw specific shapes; spatial visualization is incomplete. • Children’s OBGET data collection steps and pre-processing are presented. • Children’s OBGET samples are labeled by the proposed semi-automatic labeling.
Steps to reproduce
In order to produce a comprehensive and complete dataset of a strong psychological drawing test that detects a wide range of mental disorders is the Original Bender Gestalt drawing test. The OBGET for children is a hand sketch drawing test dataset by children aged from 4 to 11 years, including 386 samples with metadata and are labeled. Each sample dataset contains 9 drawn patterns using pencil and paper. Participants include children with or without mental health and were randomly selected in a school or psychiatric clinic. The samples are stored along with the children’s metadata such as age, gender, number of children at home, presence of parents and test result that is detected by a psychologist. Also, each pattern in each sample is labeled from 0 to 8 semi-automatically. We analyzed the data and metadata using descriptive and inferential statistics.