This research employs the lattice Boltzmann method (LBM) to investigate the influence of variable thermal properties on buoyancy-driven flow in natural convection scenarios. Through comprehensive simulations spanning Rayleigh numbers (Ra) from 103 to 107, the interplay between entropy generation and heat transfer within the cavity is examined using both Boussinesq and non-Boussinesq approximations. Notable findings reveal a transition from conduction to convection dominance with increasing Ra, where convection's impact becomes pronounced. Non-Boussinesq simulations consistently yield higher entropy and Nusselt number (Nu) values due to temperature-dependent fluid properties. Moreover, entropy generation due to fluid friction surpasses heat transfer-driven entropy at a critical Ra of 106, highlighting its predominant role. This work emphasizes Ra's pivotal role in optimizing heat transfer efficiency and offers insights into the advantages of non-Boussinesq simulations for understanding and enhancing buoyancy-driven flows.