Development and validation of deep learning architectures for Kazakhstan market forecasting with integration of ESG factors and attention mechanisms

Abstract

In this article, the authors conducted a study on the application of deep learning architectures for forecasting the dynamics of the Kazakhstan market with the integration of ESG factors and attention mechanisms. It is demonstrated that hybrid CEEMDAN-LSTM models achieve high accuracy in predicting green bonds (RMSE = 0.268), outperforming traditional methods by 30–40%. Research from 2023–2025 confirms the effectiveness of multi-head attention mechanisms, which enable up to 94% accuracy in short-term forecasts. Based on the experience of emerging markets in Central Asia, it was established that about 31% of financial institutions already employ artificial intelligence technologies, while the green bond market in Kazakhstan has reached USD 441 million with the support of the AIFC Green Finance Centre.

Special attention is given to the transformation of green finance through the integration of advanced deep learning architectures with sustainable investment strategies. The authors established that transformer models are effective for analyzing long-term ESG dependencies, whereas LSTM and GRU architectures are optimal for short-term volatility forecasting. A critical condition for improving accuracy is high-quality data preprocessing and the appropriate selection of evaluation metrics (MAE, RMSE, MAPE) adapted to the specifics of financial time series.

The results obtained show that the proposed approach can serve as a scientific and practical basis for the analysis and forecasting of the stock market, opening new opportunities for the development of sustainable finance in Kazakhstan.