My project is about automated classification of vicarious trial and error (VTE), a behavior observed in rats when they pause and look around before making decisions during a spatial memory task. During delayed spatial alternation (DSA) tasks, a rat is randomly placed on one of two start arms of a plus maze, with reward delivered on alternating arms for each trial. The movements of rats are recorded as position data while they perform the task.

Since our lab don’t have a commonly agreed upon criteria for VTE classification with our maze, manual scoring of VTE with recorded behavioral data has been time consuming, requiring many people to do the same work. Thus, my mentor and I decided to make a machine learning program to achieve automated and highly efficient VTE classification.

I first produced a representative data set with trials that were manually scored and commonly agreed by our lab members. Then, my mentor and I figured out several quantifiable features of VTEs and non-VTEs based on the representative data set. My mentor and I used machine learning algorithms to let our program learn those features that separate VTEs from not VTEs and help us accomplish automatic classification of VTEs with raw behavioral data from the DSA task. Preliminary result indicates that the supervised classification by the program aligns well with manual scoring, with roughly the same degree of agreement. Thus, I am currently transiting from a fully supervised method to a semi-supervised method, which allows almost full automation and minimal manual oversight. This work will provide insights for the behavioral strategy of rats throughout learning and guide us to find the connection between VTE behavior and neural circuitry.