Lab member: Shamal Lalvani
Diagnosis of Mental Health conditions, such as Addiction, Depression and Destructive behaviors involves expensive measurements. These measurements involve neuro-imaging, psychiatric evaluations and access to previous medical records. While these techniques are fundamental and irreplaceable, they do not scale well to large populations without significant resources. This is further complicated when the condition at hand is time-sensitive and requires significant medical intervention. Along these lines, we seek to use a system, founded in behavioral models including prospect theory, mean-variance theory and hedonic deficit theory, to identify individuals with these conditions.
To illustrate some of the work, we collect a COVID-19 dataset in which participants with and without COVID take a particular behavioral task. We frame this data in behavioral economic paradigms related to decision making. Some of these involve Prospect Theory and Mean-Variance Theory. To illustrate, we show the Value Function curve in Prospect theory in Figure 1, and the analagous associated curves in Figure 2 and those from our behavioral task.
This behavioral feature space is then used to predict covid through Gaussian Process classification and Kernel Deisgn. Additionally, due to the data imabalance, we oversample the covid positive class with gaussian mixture models as a generative model through the EM algorithm. We find that a sum of a radial basis function kernel, constant kernel and rational quadratic kernel (intuitively, an infinite sum of radial basis kernels with different length scales) allows us to segregate those with Covid with considerable accuracy (i.e. a sensitivity of more than 50 percent, while maintaining an overall accuracy of about 65%) on a on a test set that resembles the prevalence of COVID in the general population, i.e. ten percent.
Team Members who work in Mental Health / Destructive Behavior / Neuro-imaging projects from the Warren Wright Adolescent Center [WWAC]: Khrystyna Stetsiv, Nicole Vike, Byoungwoo Kim and Sumra Bari, Sean Woodward
Director of the WWAC: Hans Breiter