Abstract— <p>The paper addresses the problem of improving the accuracy of models for assessing low-temperature properties, ignitability indicators, and anti-wear properties of target products from fractionating column at limited laboratory data. To solve this problem, a method of model development is proposed, which includes an algorithm of augmenting a small training sample on fractional composition data. The algorithm is distinguished by a method of selecting additional data considering the sparsity indicator, which made it possible to include the required amount of data in the training sample and, as a result, ensure improved model quality. The use of the proposed method has enabled to increase model accuracy by an average of 18% compared to known methods and by an average of 6% compared to a method based on augmentation the training sample without considering the sparsity indicator. The results are presented using examples of model development for quality indicators estimation, such as the cold filter plugging point, flash point, kinematic viscosity at 40°C, cetane number of middle distillate (diesel fraction), and flash point of the kerosene fraction from an industrial fractionating column of a hydrocracking unit.</p>