Natural disasters preparedness measures are of paramount importance in reducing fatalities, economic tolls, and health concerns for vulnerable populations. One such preparedness measure that policymakers can take is issuing an evacuation order. Our study focuses on vulnerable populations including people in low lying areas, flood evacuation zones, low income communities, sparsely populated areas, and in manufactured or mobile homes. We analyze the evacuation decision and distance of over 170,000 individuals using passively collected location-based service (LBS) data in Florida before, during and after Hurricane Irma. We control for evacuation order, evacuation order date, and storm path. The results of the analysis are in line with hurricane evacuation studies using survey data, which provides validation that using big data can be a reliable and efficient method to analyze hurricane evacuation decision making. Our major findings are that (1) individuals issued a mandatory order are far more likely to evacuate than if given a voluntary order, (2) manufactured and mobile home residents are more likely to evacuate short distances than site-built home residents but less likely to evacuate longer distances, (3) people living in low income census tracts were less likely to evacuate longer distances than those in higher income census tracts and (4) population, employment and road density were positively associated with decision to evacuate. The authors recommend that policies and hurricane preparation measures pay particular attention to less densely populated and low-income regions as these were identified as vulnerable areas that were less likely to evacuate.