The mass composition of the Telescope Array surface detector data is analyzed with the use of the Boosted Decision Trees (BDT) technique, which assigns one parameter $\xi$ for each event, which is then concequently used for the determination of the average atomic mass of the primary particle. Itself, $\xi$ parameter may be distributed unevenly across the sky and indicate the mass composition anisotropy of the ultra-high-energy cosmic rays. The sensitivity of the BDT-based mass composition technique to the possible anisotropies is tested with the use of the $\xi$ distribution for the proton and iron Telesope Array surface detector Monte-Carlo sets.