Preprocessing Weights

One of the key pre-processing steps involves bringing the values of all columns of X to a comparable range. The neural network models do not perform well when data is highly skewed within a certain band. The data standardization process is also a security measure to prevent us from extrapolating what the original data sources could be.

In order for this data standardization for the input X, we encourage you to use the standardization formula: (X-A)/b to bring the value of X into an acceptable range. Please do not disclose the values of A and b to us. These values have to be secure with you as you will use them for consistent results when the model is deployed in production.

Please note that if A = µ and b = σ, then this is essentially a normalization operation. The data need not be normalized but it is required to be within comparable range for easier modeling operations. This consideration extends to quantitative information attributes about image data as well.

(Example:)