ARCHIVES
VOL. 12, ISSUE 9 (2025)
Risk-sensitive neural Search: A metaheuristic-enhanced LSTM architecture with volatility-aware clustering for stock price prediction
Authors
Lim Eng Aik
Abstract
This paper proposed a novel risk-sensitive
neural search framework for stock price prediction that integrates
metaheuristic optimization with volatility-aware clustering to enhance the
robustness of LSTM-based forecasting. The proposed method addresses the
limitations of conventional approaches by explicitly modeling market regimes
and incorporating risk constraints during neural architecture research. Our
hybrid architecture consists of three synergistic components: a hierarchical
clustering module that partitions historical data into volatility regimes, a
metaheuristic search module that optimizes LSTM configurations under
conditional value-at-risk constraints, and a dynamic ensemble that adaptively
combines regime-specific predictions. The clustering module employs a learnable
distance metric to capture nonlinear relationships between returns, volatility,
and macroeconomic indicators, while the neural search module balances
prediction accuracy with tail risk minimization through a temperature-annealed
selection mechanism. Furthermore, the LSTM ensemble dynamically adjusts its
weighting scheme based on both cluster assignments and macroeconomic conditions,
enabling context-aware predictions. Experimental results demonstrate
significant improvements in risk-adjusted performance metrics compared to
baseline models, particularly during high-volatility periods. The framework’s
modular design facilitates seamless integration with existing data
preprocessing and feature engineering pipelines, making it adaptable to diverse
financial datasets. This work advances the field of neural financial
forecasting by introducing a principled approach to joint optimization of
predictive accuracy and risk sensitivity, offering practical value for
quantitative trading and portfolio management applications.
Download
Pages:12-18
How to cite this article:
Lim Eng Aik "Risk-sensitive neural Search: A metaheuristic-enhanced LSTM architecture with volatility-aware clustering for stock price prediction". International Journal of Multidisciplinary Research and Development, Vol 12, Issue 9, 2025, Pages 12-18
Download Author Certificate
Please enter the email address corresponding to this article submission to download your certificate.