Scholarlink

<p>SentiVol-GA is an evolvable hybrid framework for stock price forecasting. This model integrates statistical models, deep learning architectures, financial sentiment analysis, and volatility-aware optimization through a Genetic Algorithm (GA). This framework combines five predictive models—Linear Regression, LSTM, GRU, Bi-LSTM, and ARIMA—with sentiment signals extracted from FinBERT, VADER, and the Loughran–McDonald dictionary. Its core innovation lies in a volatility-scaling mechanism that adjusts sentiment impact during turbulent periods, while the GA periodically optimizes model and sentiment weights to minimize forecasting error thus sustaining the model’s predictive stability. This empirical evaluation across eight Indian IT-sector stocks—covering large, mid- and small-cap segments, demonstrates up to 12% improvement in <i>R</i><sup>2</sup>, 30–60% reduction in RMSE, and 20–35% gain in tolerance-based accuracy over classical and deep-learning baselines. Nonparametric Friedman and Wilcoxon tests (<i>p</i> &lt; 0.05) confirm that these performance gains are statistically significant. SentiVol-GA offers a volatility-aware, sentiment-driven, and dynamically optimized approach that is scalable, interpretable, and robust for intelligent financial decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SentiVol-GA: a volatility-scaled genetic fusion of predictive models and financial sentiment for adaptive stock forecasting

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Monika Singh,
  • Harinandan Tunga,
  • Samarjit Kar

摘要

SentiVol-GA is an evolvable hybrid framework for stock price forecasting. This model integrates statistical models, deep learning architectures, financial sentiment analysis, and volatility-aware optimization through a Genetic Algorithm (GA). This framework combines five predictive models—Linear Regression, LSTM, GRU, Bi-LSTM, and ARIMA—with sentiment signals extracted from FinBERT, VADER, and the Loughran–McDonald dictionary. Its core innovation lies in a volatility-scaling mechanism that adjusts sentiment impact during turbulent periods, while the GA periodically optimizes model and sentiment weights to minimize forecasting error thus sustaining the model’s predictive stability. This empirical evaluation across eight Indian IT-sector stocks—covering large, mid- and small-cap segments, demonstrates up to 12% improvement in R2, 30–60% reduction in RMSE, and 20–35% gain in tolerance-based accuracy over classical and deep-learning baselines. Nonparametric Friedman and Wilcoxon tests (p < 0.05) confirm that these performance gains are statistically significant. SentiVol-GA offers a volatility-aware, sentiment-driven, and dynamically optimized approach that is scalable, interpretable, and robust for intelligent financial decision-making.