We develop a machine learning framework for predicting the density, viscosity, and heat capacity of a family of anion-functionalized ionic liquids for CO2 capture, specifically those with tetraalkylphosphonium cations and aprotic N-heterocyclic anions (AHAs). We screen several feature sets using group contribution-based (GC) descriptors and descriptors extracted from COSMO-RS sigma profiles (SP) to build Support Vector Regression (SVR) and Gradient-Boosted Regression (GBR) machine learning models. Viscosities and densities were modeled based on data sets containing nearly 60 ILs each. The best fit for viscosity used GC-based descriptors and the SVR model, achieving a test set %AARD of 12.5% and R2 of 0.989. Density was modeled using these same descriptors with the SVR model framework and was fitted with a test set %AARD of 1.0%. Heat capacity was fit as a function of molar volume and temperature, a general trend observed for all ILs in a family. Heat capacity predictions could then be made using the density SVR model with a test set accuracy of 3.0 %AARD. With these results, we have developed predictive models which can potentially be used in the design of new advanced ionic liquids for carbon capture.