Using Convolutional Neural Networks for the Automated Scoring of the Bender-Gestalt Test Ii
School Name
South Carolina Governor's School for Science & Mathematics
Grade Level
12th Grade
Presentation Topic
Computer Science
Presentation Type
Mentored
Abstract
This research was done as a way to simplify and improve the efficiency of the grading system of a widely used psychological test called the Bender-Gestalt test. The current hand-scored system of grading is subjective and inefficient, so we decided to come up with a way to use artificial intelligence to improve it. First, we created a data set of drawings from the test that we could use to train a network on. Then we created code for a convolutional neural network, which is used in image processing/ identification software, and fed it our data set. After it trained on the data, it came back with a 99% accuracy rating, meaning our project was highly successful. The next step would be to gain access to real data from the testing company and if our code is successful there too, we could sell our code to the company.
Recommended Citation
Harrington, Caitlin and Ramsey, William, "Using Convolutional Neural Networks for the Automated Scoring of the Bender-Gestalt Test Ii" (2019). South Carolina Junior Academy of Science. 296.
https://scholarexchange.furman.edu/scjas/2019/all/296
Location
Founders Hall 140 A
Start Date
3-30-2019 9:45 AM
Presentation Format
Oral Only
Group Project
Yes
Using Convolutional Neural Networks for the Automated Scoring of the Bender-Gestalt Test Ii
Founders Hall 140 A
This research was done as a way to simplify and improve the efficiency of the grading system of a widely used psychological test called the Bender-Gestalt test. The current hand-scored system of grading is subjective and inefficient, so we decided to come up with a way to use artificial intelligence to improve it. First, we created a data set of drawings from the test that we could use to train a network on. Then we created code for a convolutional neural network, which is used in image processing/ identification software, and fed it our data set. After it trained on the data, it came back with a 99% accuracy rating, meaning our project was highly successful. The next step would be to gain access to real data from the testing company and if our code is successful there too, we could sell our code to the company.
