The operation of a modern electrical system presents complex operating conditions, and uncertainty factors increased by the presence of variable and flexible loads. This stochastic operating point variation can promote low-frequency oscillations, and like the conventional power system stabilizer (PSS) features static tuning, it does not provide sufficient damping. Today, an electrical system is monitored through online measurements from the wide-area measurement system (WAMS); moreover, it enables stochastic dynamics to be tracked as the remote signals of WAMS system, the time delay due to the communication channel. It is interesting to analyze the incorporation of the PSS of signals from the WAMS without the time delay effect that degrades the small-signal stability (SSS) and, in turn, allows adaptive tuning. Through two stages, at first, the subspace of the operating point is determined based on a distribution of consumption levels. Each operating point considered a random time delay. For each subspace, establish a bank of tuned PSSs. Second, through the optimal classification and the regression decision tree (CART), the classification rules for the subspace are determined, allowing classifying the measurements of WAMS. The proposed methodology is applied to the New England of 66-bus system and Ecuadorian electric system to demonstrate that the adaptive tuning of PSSs significantly improves SSS under different scenarios and is probabilistic robust, considering the time delay.