A binding stability order predicting, molecular docking based fast technique was developed for host--guest complexes. Molecular descriptors were applied to ligand molecules to make the binding energy based docking scoring functions more efficient and reach the $\pm $0.50 log K unit theoretical precision for predictions. The goal of this work was to model complexes of cucurbit[7]uril (CB7) as a host molecule with different local anesthetics and choline and phosphonium choline molecules. The guest molecules were docked into the cavity of CB7. The binding free energy maps were correlated by Contact, Grid-Based, Hawkins GB/SA, and AMBER docking scores obtained from UCSF DOCK software. The guest molecules were clustered by the scores. Linear correlations on the ln K - docking score plots were characterized on the learning set and log K was reproduced on the test set within an accuracy of $\pm $0.92 log K units.