Increasing temperature and changing precipitation due to warming climate may increase wildfire activities over the contiguous US (CONUS). Prior studies have projected increased fire activities in CONUS in the future, but different approaches and future scenarios yield a wide range of estimates and lack a full understanding of key factors driving future fire activity and associated uncertainties. Here, we construct a neural network (NN) model explained by the Shapley Additive explanation (SHAP) to predict fire PM_2.5 emissions and to understand factors driving changes in fire emissions over CONUS in the mid-21^st century under a high greenhouse gas emissions scenario (SSP5-8.5). Using meteorology and leaf area index (LAI) simulated by eight global climate models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) as input to the NN model, the total fire PM_2.5 emissions are projected to increase by 4-75% (model spread) over CONUS. Among different regions, fire emissions in the southwestern US and western forest region are projected to increase more significantly by 27-154% and 11-196%, respectively. The largest enhancement occurs in June-July-August (JJA) over western US, with the median ratios of future to present-day fire emissions ranging from 1.67 to 2.86, which are mainly driven by increasing temperature (2-4 K) and decreasing soil moisture (2-10%) in the future. Considering the future land use and land cover change simulated by four of the eight global climate models with biogeochemical cycles, the projected fire emissions increase by 58%-83% over western US, compared to the future projection without land use and land cover change, which is driven by future increases in vegetation fraction (15%). The results highlight the important role of the drying trend resulting from warmer temperature and land use and land cover change in increasing fire emissions in the future.