Irina V. Manakhova, Aleksandr V. Matytsyn Innovative Approaches to Training Data Generation for Oil Demand Forecasting

DOI: 10.38050/2078-3809-2025-17-4-9-34

Abstract

This study explores methods for generating training data to improve demand forecasting accuracy in the oil market. The limitations of traditional approaches are examined, and the use of generative adversarial networks, specifically the TimeGAN (Time-series Generative Adversarial Network) model, is proposed for creating synthetic time series data. The results demonstrate that TimeGAN can generate realistic data closely resembling actual data, preserving market volatility and structural characteristics. However, model limitations were identified, suggesting the need for further research to enhance forecast efficiency and accuracy on the oil in volatile market conditions.

Keywords: deep learning, generative adversarial networks, training data, model TimeGAN.

JEL: Е17, С53.

For citation: Manakhova, I.V., Matytsyn, V.M. (2025) Innovative Approaches to Training Data Generation for Oil Demand Forecasting. Scientific Research of Faculty of Economics. Electronic Journal, vol. 17, no. 4, pp. 9-34. DOI: 10.38050/2078-3809-2025-17-4-9-34.

Download