Embodiments of the invention are directed to systems, methods, and computer program products for utilizing machine learning to identify data which is to be obfuscated in a format-preserving manner, which allows the obfuscated or masked data to appear as though it is original data. Because this type of obfuscation technique may require a higher degree of computational power than other techniques, there is a need to be able to dynamically choose when to implement format preservation based on a variety of factors. By using machine learning techniques, the present invention provides the functional benefit of analyzing both the data to be obfuscated, as well as available computational resources, to determine when it is appropriate to apply a format-preserving masking algorithm to the data. Accordingly, the present invention may ensure that organizational data is appropriately masked while preventing the resource strain associated with preserving the format of all original data.