Access to accurate estimates of water withdrawal is requisite for urban planners as well as operators of critical infrastructure systems to make optimal operational decisions and investment plans to ensure reliable and affordable provisioning of water. Furthermore, identifying the key predictors of water withdrawal is important to regulators for promoting sustainable development policies to reduce water use. In this paper, we developed a rigorously evaluated predictive model, using statistical learning theory, to estimate state-level, per-capita water withdrawal as a function of various geographic, climatic and socio-economic variables. We then harnessed the data-driven predictive model to identify the key factors associated with high water-usage intensity among different sectors in the U.S. We analyzed the predictive accuracy of a range of parametric models (e.g., generalized linear models) and non-parametric, flexible learning algorithms (e.g., generalized additive models, multivariate adaptive regression splines and random forest). Our results identified irrigated farming, thermo-electric energy generation and urbanization as the most water-intensive anthropogenic activities, on a per-capita basis. Among the climate factors, precipitation was also found to be a key predictor of per-capita water withdrawal, with drier conditions associated with higher water withdrawals. Results of the first-order sensitivity analysis indicated changes between +/-10% in the future water withdrawal across the U.S., in response to precipitation changes, by the end of the 21st Century under the business-as-usual scenario. Overall, our study highlights the utility of leveraging statistical learning theory in developing data-driven models that can yield valuable insights related to the water withdrawal patterns across expansive geographical areas.