One of the uncertainties in interpretation of ultra-high energy cosmic ray (UHECR) data comes from the hadronic interaction models used for air shower Monte Carlo (MC) simulations. The number of muons observed at the ground from UHECR-induced air showers is expected to depend upon the hadronic interaction model. One may therefore test the hadronic interaction models by comparing the measured number of muons with the MC prediction. In this paper, we present the results of studies of muon densities in UHE extensive air showers obtained by analyzing the signal of surface detector stations which should have high $\it{muon \, purity}$. The muon purity of a station will depend on both the inclination of the shower and the relative position of the station. In 7 years' data from the Telescope Array experiment, we find that the number of particles observed for signals with an expected muon purity of $\sim$65% at a lateral distance of 2000 m from the shower core is $1.72 \pm 0.10{\rm (stat.)} \pm 0.37 {\rm (syst.)}$ times larger than the MC prediction value using the QGSJET II-03 model for proton-induced showers. A similar effect is also seen in comparisons with other hadronic models such as QGSJET II-04, which shows a $1.67 \pm 0.10 \pm 0.36$ excess. We also studied the dependence of these excesses on lateral distances and found a slower decrease of the lateral distribution of muons in the data as compared to the MC, causing larger discrepancy at larger lateral distances.