Non Normal Distribution Transformations in Continuous Reactor Scale Up
Applying Box-Cox and Johnson transformations to skewed residence time data prevents major capability miscalculations during continuous reactor scale up.
Yeo-Johnson Algorithm
Meaning
Power transformation method designed to stabilize variance and make data more closely resemble a normal distribution. The yeo-johnson algorithm is an extension of the Box-Cox method that can handle both positive and negative values. It applies different mathematical functions depending on whether the input is greater than or equal to zero.
This versatility makes it a standard tool for preprocessing features in predictive modeling and quality control analysis.
Parameter Choice
Selection of the lambda value determines the strength of the transformation. The yeo-johnson algorithm uses maximum likelihood estimation to find the optimal power. This parameter captures the specific skew of the distribution.
Mathematical Consistency
Continuity is maintained at the zero point across all possible values of the power parameter. In the yeo-johnson algorithm, the formulas for positive and negative inputs are constructed to meet at a single point with a continuous derivative. This ensures that the transformed scale is smooth and does not introduce artificial gaps in the distribution.
It allows for the analysis of zero centered noise in sensor data.
Predictive Accuracy
Normalizing inputs reduces the influence of extreme outliers on the final model. Applying the yeo-johnson algorithm often improves the performance of linear regression and other algorithms that assume a Gaussian error structure. It is particularly useful in manufacturing where process variables can span multiple orders of magnitude.
Consistent scaling across features leads to more stable and interpretable results.