Neural stimulation with focused ultrasound is a rapidly developing research field. This thesis demonstrates design and validation of a comprehensive focused ultrasound neural stimulation (FUS-NS) research system that can both generate ultrasound pulses with user-definable parameters and process the elicited Electroencephalographic (EEG) responses from human test subjects. Several ultrasound transducers were characterized with different methods including both simulation and experimental measurements to ensure safety and confirm parameter space. Parameter sets were swept in acoustic tank and measured with hydrophone before human subject tests to ensure safety. EEG data were analyzed with custom-built MATLAB script to sort and extract user-definable patterns. Different versions of the transmit sub-system (TX) was built and tested and will be presented in this thesis. Our human study showed that FUS-NS system was safe for use in human subjects and activated peripheral neural nociceptors (free nerve endings), and a resultant painful sensation was perceived when the system was tested. The pain sensation was recorded with our NR EEG evoked potentials (EP). The human subject study (N=3) carried out confirmed safety of the system as well as all the proposed functionalities. This system will greatly facilitate the research in focused ultrasound neural stimulation. Our laboratory has now moved to characterize additional FUS-NS aimed to stimulate axonal neuronal structures.

Fraud detection is a problem of highly imbalanced binary classification, where the size ratio of positive class (fraudulent instances) to negative class (normal instances) is low. Any fraud detection model is faced with the trade-off between two types of classification mistakes - type I error (also known as the false positive rate (FPR)) and type II error (also known as the false negative rate (FNR)). One common way to handle this trade-off is to set a pre-determined level of maximum acceptable type I error, $\alpha$, and try to minimize type II error. However, an acceptable empirical type I error evaluated on the test set (below $\alpha$) does not necessarily guarantee population type I error to be below $\alpha$ as well. This problem is addressed by the Neyman-Pearson (NP) classification algorithm \citep{np}, with which we can put a high-probability upper bound on population type I error. In this study, we use simulation to illustrate how empirical and population type I error can differ, and how NP classification effectively controls population type I error. We then build credit card fraud prediction models based on classical and NP algorithms and compare their performances.

During the past two to three decades, and especially during the Covid-19 pandemic, e-shopping has become increasingly popular, changing the way people shop and travel. With increasing concerns about the environmental impacts of transportation, particularly on regional air quality and on emissions of greenhouse gases (GHG), it is important to understand how e-shopping has affected household travel behavior.In this dissertation, I investigated the influence of e-shopping before, during, and after the pandemic by analyzing data from the 2009 and the 2017 U.S. National Household Travel Surveys (NHTS), from the 2017 American Time Use Survey (ATUS), and from an IPSOS survey of Californians conducted in late May 2021. Understanding changes in shopping is essential to business owners, logistics managers (for adapting supply chains), transportation planners (for mitigating the impacts of warehousing and of additional residential freight deliveries), and policymakers (for helping at-risk and underserved groups). This dissertation has three parts. In the first part, I estimated zero-inflated negative binomial models to analyze factors that affected residential deliveries before the pandemic based on the 2009 and 2017 NHTS. I found that e-shoppers in the U.S. were more varied in 2017 than in 2009. Households with more females, higher incomes, and more education, received more deliveries. I also analyzed the 2017 ATUS to explore factors that influence grocery shopping. I found that in-store grocery shoppers were more likely to be female and unemployed but less likely to be younger, to have less than a college education, and to be African American. In contrast, online grocery shoppers were more likely to be female. In the second part, I studied the impact of e-shopping on household travel using propensity score matching. My analysis of 2017 NHTS data showed that before the pandemic, greater online shopping was associated with more frequent trips and slightly more travel. Furthermore, the extent to which an increase in the number of activities translated into more travel depends on population density, the day of the week, the frequency of online shopping, and the type of activity. In the third part, I analyzed the impact of the Covid-19 pandemic on grocery shopping frequency in-store, and online with home delivery (e-grocery) or pickup (click-and-pick), to understand how they changed due to the pandemic, and how they may change after, using ordered models and structural equation models. My results showed that Californians kept shopping for groceries in brick-and-mortar stores during the pandemic but less frequently than before. The pandemic accelerated the adoption of e-grocery and click-and-pick with some strong generation effects: younger generations were more likely to experiment with e-grocery and click-and-pick, while older generations relied more on in-store shopping. Education also made a difference, but thankfully race did not impact the use of e-grocery and click-and-pick, and intentions to use e-grocery and click-and-pick (but it did affect in-store grocery shopping before). My results also illustrated the heterogeneity of Hispanics. As expected, tech-savvy households were much more likely to embrace e-grocery and click-and-pick.