The results on ultra-high energy cosmic rays' chemical composition based on the data from the Telescope Array surface detector are reported. The Telescope Array (TA) is an experiment, located in Utah, USA, designed for observation of extensive air showers from the ultra-high energy cosmic rays. TA surface detector (SD) array consists of 507 detector stations, placed in a square grid with 1.2\,km spacing with w total area of approximately 700 $\mbox{km}^2$. Each station has two layers of 1.2\,cm thick plastic scintillator of 3 $\mbox{m}^2$ area. This talk is focused on the analysis of the mass composition of primary particles based on the 9 years data of the TA surface detector. The method employs the Boosted Decision Trees (BDT) technique for the analysis of the multiple composition-sensitive shower parameters. The BDT classifier is trained with the two Monte-Carlo training sets: proton, which is considered as background events, and iron, considered as signal events. The classifiers results in a single variable $\xi$ for data and Monte-Carlo test sets. The data to Monte-Carlo comparison results in an average atomic mass of UHECR for energy range $10^{18.0}-10^{20.0}\ \mbox{eV}$. The comparison with TA hybrid composition results and the other experiments is presented.