Outliers frequently occur during data collection, introducing deviations that can significantly degrade system identification performance. To address this issue, an adaptive suppression mechanism based the expectation-maximization algorithm is deduced. By incorporating a penalty on the outlier vector and assigning individual weights to each data point, the algorithm ensures that outlier weights are lower than normal samples. Consequently, outliers are excluded from the contaminated data without requiring prior knowledge, thus minimizing estimation bias. The deduced algorithm is validated through a mass-spring-damper system.