The CMA-ES-PDM is a variant of the CMA-ES designed for solving mixed-integer black-box optimization problems. In this method, elite points generated during the evolution process are selected to build sequential probability distribution models for the integer variables. These models are then partially used to generate the integer parts of candidate solutions in some next iterations. The performance of this method depends on a marginal probability, denoted by \(\alpha \) , which must be carefully selected for each objective function. A higher \(\alpha \) value typically results in a higher success rate but slower convergence. Conversely, a lower \(\alpha \) value may lead to stagnation and a lower success rate. In this work, we propose a method to choose the value of \(\alpha \) randomly in order to balance the exploration and exploitation of the algorithm. Additionally, the probability model is smoothed when certain intermediate values have zero probability, enhancing the sampling process for integer components. Numerical experiments on the MI-BBO benchmark problems demonstrate that these modifications improve both success rate and convergence speed on certain test functions.