Scaling tabular data of training datasets with neural networks

Abstract

The paper proposes a method of scaling the tabular data of the training dataset using neural networks, describes the architecture of such networks.

Relevance. Presently, there is a problem of insufficient amount of raw data for training artificial intelligence models, which leads to significant modeling error. The work is devoted to the development of approaches to the generation of artificial tabular data, which can be used in the future for artificial intelligence models.

Goal. The purpose of the work was to analyze methods and algorithms for scaling the training dataset for tabular data using neural networks.

Research methods. The main research method is the process of selecting the parameters of the artificial data generation algorithm and choosing the optimal parameters of the neural network architecture.

The results. Using neural networks for scaling the tabular data of the training dataset confirmed the efficiency of the proposed approach. The results of the algorithm adjustment and the selection of the optimal parameters of the neural network showed that the generated artificial data most resemble the initial ones in terms of the criteria of average value, maximum, minimum and dependence between data.

Conclusions. The task of scaling the tabular data of the training dataset using neural networks has been solved. This approach makes it possible to significantly simplify the process of learning neural networks. The scientific novelty of this work lies in the development of approaches and methods for increasing tabular data using artificial intelligence and deep learning.