This study explores the effect of different phases of unconventional shale gas well-pad development on ambient air quality and the relationship between ambient concentrations of air pollutants and operator activity. The U.S. Department of Energy's National Energy Technology Laboratory operated a mobile air-monitoring laboratory on two shale well pad sites in Pennsylvania and six shale well pad sites in West Virginia. The purpose of this study is to integrate expert knowledge and collected ambient air monitoring data by developing a Bayesian network (BN) model. The monitoring period included well-pad site development; construction, including vertical and horizontal drilling; hydraulic fracturing; flowback; and production. The observed data includes meteorological data with high time resolution and air quality data (volatile organic compounds (VOCs), ozone, methane and carbon isotopes in methane, carbon dioxide (CO2) and carbon isotopes in CO2, coarse and fine particulate matter (PM10 and PM2.5), and organic and elemental carbon). The results provide useful information for evaluating the influence of on- and off-site pollutant sources and determining future research efforts for building the BN model. The overall results of the developed six scenarios show that the prediction power of the proposed model for the vertical drilling phase is 94%. The high concentration of methane increases the probability of fracturing phase as source; the low concentration of PM10 and O-3 occurrence increases the same probability to 82%; the low concentration of ethane and CO2 increases the probability to 98%. This study shows how expert Bayesian models can improve our ability to predict future air pollution risk associated with unconventional shale gas development.