When using regression models, inputs with highly skewed distributions can lead to:

Using regression models with inputs that have highly skewed distributions can lead to bias in predictions. This occurs because many regression techniques, particularly those based on ordinary least squares (OLS), assume that the relationships between predictors and the response variable are linear and that the input data follows a normal distribution. When the distribution of inputs is skewed, the assumptions of linearity and normality can be violated, which often results in the model being unable to accurately capture the actual relationships within the data. Consequently, this can produce biased coefficient estimates and unreliable predictions, as the model may unduly emphasize or neglect certain ranges of the input variable.

In contrast, improved prediction reliability typically arises from well-distributed input variables, while skewed data can contribute to complications in understanding the relationships modeled. Regarding interpretability, a model's complexity is generally increased by handling skewed distributions, requiring transformations or different modeling approaches rather than being streamlined. Therefore, when working with regression models, addressing skewed distributions is critical to ensuring that predictions are valid and reliable.